Dogs, Rush hour traffic, and the history of storage I/O benchmarking–Part 2

Part one of "History of storage I/O Benchmarking" attempted to demonstrate how Synthetic Benchmarks on their own simply cannot generate or measure storage performance in a meaningful way. Emulating real workloads seems like such a simple matter to solve, but in truth it is a complex problem involving technical, and non-technical challenges.

  • Assessing workload characteristics is often left for conjecture.  Understanding the correct elements to observe is the first step to simulating them for testing, but how the problem is viewed is often left for whatever tool is readily available.  Many of these tools may look at the wrong variables.  A storage array might have great tools for monitoring the array, but is an incomplete view as it relates to the VM or application performance.
  • Understanding performance in a Datacenter crosses boundaries of subject matter expertise.  A traditional Storage Administrator will see the world in the same way the array views it.  Blocks, LUNS, queues, and transport protocols.  Ask them about performance and be prepared for a monologue on rotational latencies, RAID striping efficiencies and read/write handling.  What about the latency as seen by the VM?  Don’t be surprised if that is never mentioned.  It may not even be their fault, since their view of the infrastructure may be limited by access control.
  • When introducing a new solution that uses a technology like Flash, the word itself is seen as a superlative, not a technology.  The name implies instant, fast, and other super-hero like qualities.  Brilliant industry marketing, but it comes at a cost.  Storage solutions are often improperly tested after some technology with Flash is introduced because conventional wisdom says it is universally faster than anything in the past.  A simplified and incorrect assertion.

Evaluating performance demands a balance of understanding the infrastructure, the workloads, and the software platforms they run on. This takes time and the correct tools for insight – something most are lacking. Part one described the characteristics of real workloads that are difficult to emulate, plus the flawed approach of testing in a clustered compute environment. Unfortunately, it doesn’t end there. There is another factor to be considered; the physical characteristics of storage performance tiering layers, and the logic moving data between those layers.

Storage Performance tiering
Most Datacenters deliver storage performance using multiple persistent storage tiers and various forms of caching and buffering. Synthetic benchmarks force a behavior on these tiers that may be unrealistic. Many times this is difficult to decipher, as the tier sizes and data handling can be obfuscated by a storage vendor or unknown by the tester. What we do know is that storage tiering can certainly come in all shapes and sizes. Whether it traditional array with data progression techniques, a hybrid array, a decoupled architecture like PernixData FVP, or a Hyper Converged solution. The reality is that this tiering occurs all the time

With that in mind, there are two distinct approaches to test these environments.

  • Testing storage in a way to guarantee no I/O data comes from and goes to a top performing tier.
  • Testing storage in a way to guarantee that all I/O data comes from and goes to a top performing tier.

Which method is right for you? Both methods are neither right nor wrong as each can serve a purpose. Let’s use the car analogy again

  • Some might be averse to driving an electric car that only has a 100 mile range.  But what if you had a commute that rarely ever went more than 30 miles a day?  Think of that as like a caching/buffering tier.  If a caching layer is large enough that it might serve that I/O 95% of the time, well then, it may not be necessary to focus on testing performance from that lower tier of storage. 
  • In that same spirit, let’s say that same owner changed jobs and drove 200 miles a day.  That same car is a pretty poor solution for the job.  Similarly, if a storage array had just 20GB of caching/buffering for 100TB of persistent storage, the realistic working set size of each of the VMs that live on that storage would realize very little benefit from that 20GB of caching space.  In that case, it would be better to test the performance of the lower tier of storage.

What about testing the storage in a way to guarantee that data comes from all tiers?  Mixing a combination of the two sounds ideal, but often will not simulate the way real data will reside on the tiers, and produces a result that is difficult to determine if it reflects the way a real workload will behave. Due to the lack of identifying these caching tier sizes, or no true way to isolate a tier, this ironically ends up being the approach most commonly used – by accident alone.

When generating synthetic workloads that have a large proportion of writes, it can often be quite easy to hit buffer limit thresholds. Once again this is due to a benchmark committing every CPU cycle as a write I/O and for unrealistic periods of time. Even in extremely write intensive environments, this is completely unrealistic. It is for that reason that one can create a behavior with a synthetic benchmark against a tiered storage solution that rarely, if ever, happens in a real world environment.

When generating read I/O based synthetic tests using a large test file, those reads may sometimes hit the caching tier, and other times hit the slowest tier, which may show sporadic results. The reaction to this result often leads to running the test longer. The problem however is the testing approach, not the length of the test. Understanding the working set size of a VM is key, and should dictate how best to test in your environment. How do you determine a working set size? Let’s save that for a future post. Ultimately it is real workloads that matter, so the more you can emulate the real workloads, the better.

Storage caching population and eviction. Not all caching is the same
Caching layers in storage solutions can come in all shapes and sizes, but they depend on rules of engagement that may be difficult to determine. An example of two very important characteristics would be:

  • How they place data in cache.  Is some sort of predictive "data progression" algorithm being used?  Are the tiers using Write-Through caching to populate the cache in addition to population from data fetched from the backend storage. 
  • How they choose to evict data from cache.  Does the tier use "First-in-First-Out" (FIFO), Least Recently Used (LRU), Least Frequently Used (LFU) or some other approach for eviction?

Synthetic benchmarks do not accommodate this well.  Real world workloads will depend highly on them however, and the differences show up only in production environments.

Other testing mistakes
As if there weren’t enough ways to screw up a test, here are a few other common storage performance testing mistakes.

  • Not testing as close to the application level as possible.  This sort of functional testing will be more in line with how the application (and OS it lives on) handles real world data.
  • Long test durations.  Synthetic benchmarks are of little use when running an exhaustive (multi-hour) set of tests.  It tells very little, and just wastes time.
  • Overlooking a parameter on a benchmark.  Settings matter because they can yield very different results.
    Misunderstanding the read/write ratios of an environment.  Are you calculating your ratio by IOPS, or Throughput?  This alone can lead to two very different results.
  • Misunderstanding of typical I/O sizes in organization for reads and writes.  How are you choosing to determine what the typical I/O size is?
  • Testing reads and writes like two independent objectives.  Real workloads do not work like this, so there is little reason to test like this.
  • Using a final ‘score’ provided by a benchmark.  The focus should be on the behavior for the duration of the test.  Especially with technologies like Flash, careful attention should be paid to side effects from garbage collection techniques and other events that cause latency spikes. Those spikes matter.

Testing organizations often are vying for a position as a testing authority, or push methods or standards that somehow eliminate the mistakes described in this blog post series. Unfortunately that is not the case, but it does not matter anyway, as it is your data, and your workloads that count.

Making good use of synthetic benchmarks
It may come across that Synthetic Benchmarks or Synthetic Load Generators are useless. That is untrue. In fact, I use them all the time. Just not the way they conventional wisdom indicates. The real benefit comes once you accept the fact that they do not simulate real workloads. Here are a few scenarios in which they are quite useful.

  • Steady-state load generation.  This is especially useful in virtualized environments when you are trying to create load against a few systems.  It can be a great way to learn and troubleshoot.
  • Micro-benchmarking.  This is really about taking a small snippet of a workload, and attempting to emulate it for testing and evaluation.  Often times the test may only be 5 to 30 seconds, but will provide a chance to capture what is needed.  It’s more about generating I/O to observe behavior than testing absolute performance.  Look here for a good example.
  • Comparing independent hardware components.  This is a great way to show differences an old and new SSD.
  • Help provide broader insight to the bigger architectural picture.

Observe, Learn and Test
To avoid wasting time "testing" meaningless conditions, spend some time in vCenter, esxtop, and other methods to capture statistics. Learn about your existing workloads before running a benchmark. Collaborating with an internal application owner can make better use of your testing efforts. For instance, if you are looking to improve your SQL performance, create a series of tests or modify an existing batch job to run inside of SQL to establish baselines and observe behavior. Test at the busiest time and the quietest time of the day, as they both provide great data points. This approach was incredibly helpful for me when I was optimizing an environment for code compiling.

Try not to lose sight of the fact that testing storage performance is not about testing an array. It’s about testing how your workloads behave against your storage architecture. Real applications always tell the real story. The reason why most dislike this answer is that it is difficult to repeat, and challenging to measure the right way. Testing the correct way can mean you might spend a little time better understanding the demand your applications put on your environment.

And here you thought you ran out of things to do for the day. Happy testing.







Your Intel NUC Home Lab questions answered

With my recent post on what’s currently running in my vSphere Home Lab, I received a number of questions about one particular part of the lab; that being my Management Cluster built with Intel NUCs. So here is a quick compilation of those questions (with answers) I’ve had around this topic.

Why did you go with a NUC?  There are cheaper options.
My approach for a Home Lab Management Cluster was a bit different than my regular Lab Cluster. I wanted to take a minimalist approach, and provide just enough resources to get my primary VMs off of my other two hosts that I do a majority of my testing against. In other words, less is more. There is a bit of a price premium with a NUC, but there also is a distinct payoff with them that often gets overlooked. If they do not keep up with your needs in the Home Lab, they can be easily repurposed as a workstation or a media PC. The same can’t be said for most Home Lab gear.

Is there anything special you have to do to run ESXi on a NUC?
Nothing terribly difficult. The buildup of ESXi on the NUC is relatively straightforward, and there is a growing number of posts that walk through this nicely. The primary steps are:

  1. Build a customized ISO by packing up an Intel NIC driver, and a SATA Controller driver, and place it on a bootable USB.
  2. Temporarily disable AHCI in the BIOS for just the installation process
  3. Install ESXi
  4. Re-enable AHCI in the BIOS after the installation of ESXi is complete.

How many cores?
The 3rd generation NUC is built off of the Intel Core i5-4250U (Haswell) chipset. It has two physical cores, and will present 4CPUs with Hyper-Threading. After managing and watching real workloads for years, my position on Hyper-Threading is a bit more conservative than many others. It is certainly better than nothing, but many times the effective performance gain is limited, and varies with workload characteristics. It’s primary benefit with the NUC is that you can run a 4vCPU VM if you need to. Utilization of the CPU from a cluster perspective are often hovering below 10%.

Is working with 16GB of RAM painful?
Having just 16GB of RAM might be a more visible pain if it were serving something other than Management VMs. The biggest issue with a "skinny" two node Management Cluster usually comes in when you have to throw one into maintenance mode. But much like having a single switch in a Home Lab, you just deal with it. Below is what the Memory usage on these NUCs look like.


There are a few options to improve this situation.

1. Trimming up some of your VMs might be a good start. Virtual Appliances like the VCSA are built with a healthy chunk of RAM configured by default (supporting all of that Java goodness). Here is a way to trim up memory resources on the VCSA, although, I have not done this yet because I haven’t needed to. Just don’t use the Active Memory metric as your sole data point to trim up a VM Memory configuration. See Observations with the Active Memory metric in vSphere on how easily that metric can be misinterpreted.

2. Look into a new option for increasing RAM density on the NUC. Yeah, this might blow your budget, but if you really want 32GB of RAM in a NUC, you can do it. At this time, the cost delta for making a modification like this is pretty steep, and it may make more sense to purchase a 3rd NUC for a Management Cluster.

3. Adjust expectations and accept the fact that you might have a little memory ballooning or swapping going on. This is by far the easiest, and most affordable way to go.

How is ESXi on a single NIC?
Well, there is no getting around the fact that the NUC comes with a single 1GbE NIC. This means no redundancy, and limited bandwidth. The good news is that with just one NIC, you can monitor this quite easily in vCenter!   Since you are running all services and data across a single uplink, it may be in your best interest to run a Virtual Distributed Switch (VDS) to properly control ingress and egress traffic, and make sure something like a vMotion isn’t going to wreak havoc on your environment. However, transitioning a vCenter VM to a VDS with a single uplink can sometimes be a little adventurous, so you might want to plan ahead.

If you must have a 2nd NIC to the host, take a look here. Nicholas Farmer showed quite a bit of ingenuity in coming up with a second uplink. Also, don’t forget to look at his great mini-rack he made for the NUCs out of Legos. Great stuff.

How do they perform?
Exactly the way I want them to perform. Out of sight, and out of mind.  Again, my primary lab work is performed on my Micro-ATX style hosts, so as long as the NUCs can keep the various Management and infrastructure VMs running, then that is good with me. Some VMs are easy to trim up to provide minimal resources (Linux based syslog servers, DNS, etc.) while others are more difficult or not worth the hassle.

Why did you put two SSDs in them?
This was for flexibility. I wanted one (mSATA) drive for the possibility of local storage if I decided to place any VMs locally, as well as another device (2.5" SSD) for other uses. In this case, I decided to apply a little PernixData FVP magic on them and use one of them in each host to accelerate the VMs. The image below shows the latency of the VCSA, which has about 99% writes. Note how the latency dropped after transitioning the VM from Write-Through (read caching) to Write-Back (read and write caching) to a consistently low level. Not bad considering all traffic is riding across a single link, and the flash device is an old SSD.

(click on image to enlarge)



Would you recommend them?
I think the Intel NUCs serve as a great little host for a Management Cluster in a Home Lab. They won’t be replacing my Micro-ATX style boxes any time soon, nor should they ever be part of a real environment, but they allow me the freedom to experiment and test on the primary part of the Lab, which is what a Home Lab is all about.

Thanks for reading.

– Pete

VDI for me. Part 5 – The wrap up


Now that VMware View is up and running, you might be curious to know how it is working.  Well, you’re in luck, because this post is about how View worked, and what was learned from this pilot project.  But first, here is a quick recap of what has been covered so far.

VDI for me. Part 1 – Why the interest 
VDI for me. Part 2 – The Plan
VDI for me. Part 3 – Firewall considerations
VDI for me. Part 4 – Connection Servers and tuning

I was given the opportunity to try VMware View for a few different reasons (found here).  I wasn’t entirely sure what to expect, but was determined to get a good feel for what VDI in 2012 could do.  Hopefully this series has helped you gain an understanding as well. 

The user experience
Once things were operational, the ease and ubiquity of access to the systems was impressive.  One of our most frequent users often stated that he simply forgot where the work was actually being performed.  Comments like that are a good indicator of success.  From a remote interaction standpoint, the improvements most often showed up where it was really needed; remote display over highly latent connections, with convenience of access.  Being able to access a remote system from behind one corporate network to another was as productive as it was cool. 

It was interesting to observe how some interpreted the technology.  Some embraced it for what it was (an appliance to be more productive), while others chose to be more suspicious.  You may have users who complain about their existing computers, but are apprehensive at the notion of it being taken away for something that isn’t tangible.  Don’t underestimate this emotional connection between user and computer.  It’s a weird, but very real aspect of a deployment like this. 

Virtualization Administrators know that good performance is often a result of a collection of components (storage, network, CPU, hypervisor) working well together through a good design.  Those of us who have virtualized our infrastructures are accustomed to this.  Users are not.  As VMs become more exposed to the end users (whether they be for VDI, or other user-facing needs), your technical users may become overly curious by what’s “under the hood” with their VM.  This can be a problem.  Comparisons between their physical machine and the VM are inevitable, and they may interpret a VM with half the processors and RAM as their physical machine to provide only half of the experience.  You might even be able to demonstrate that the VM is indeed better performing in many ways, yet the response might be that they still don’t have enough RAM, CPU, etc.  The end user knows nothing about hypervisors or IOPS, but they will pay attention to some of the common specifications general consumers of technology have been taught to care about; RAM and CPUs.

So in other words, there will be aspects of a project like this that have everything to do with virtualization, yet nothing to do with virtualization.  It can be as much of a people issue as it is a technical issue.

Other Impressions
The PCoIP protocol is very nice, and really shines in certain situations. I love the fact that it is a tunable, non-connection oriented protocol that leaves all of the rendering up to the host. It just makes so much sense for remote displays. But it can have characteristics that make it feel different to the end user. The old “window shake” test might redraw itself slightly different than in a native display, or using something like RDP. This is something that the user may or may not notice.

The pilot program included the trial of a PCoIP based Zero Client. The Wyse P20 didn’t disappoint. Whether it was connecting to a VM brokered by View, or a physical workstation with a PCoIP host card brokered by View, the experience was clean and easy. Hook up a couple of monitors to it, and turn it on. It finds the connection server, so all you need to do is enter your credentials, and you are in. The zero client was limited to just PCoIP, so if you need flexibility in that department, perhaps a thin client might be more appropriate for you. I wanted to see what no hassle was really like.

As far as feedback, the top three questions I usually received from users went something like this:

“Does it run on Linux?”

“How come it doesn’t run on Linux?”

“When is it going to run on Linux?”

And they weren’t just talking about the View Client (which as of this date will run on Ubuntu 11.04), but more importantly, the View Agent.  There are entire infrastructures out there that use frameworks and solutions that run on nothing but Linux.  This is true especially in arenas like Software Development, CAE and Scientific communities.  Even many of VMware’s offerings are built off of frameworks that have little to do with Windows.  The impression that the supported platforms of View gave to our end users was that VMware’s family of solutions were just Windows based.  Most of you reading this know that simply isn’t true.  I hope VMware takes a look at getting View agents and clients out for Linux.

Serving up physical systems using View as the connection broker is an interesting tangent to the whole VDI experience.  But of course, this is a one user to one workstation arrangement – its just that the workstation isn’t stuffed under a desk somewhere.  I suspect that VMware and its competitors are going to have to tackle the problem of how to harness GPU power through the hypervisor so that all but the most demanding of high end systems can be virtualized.  Will it happen with specialized video cards likely to come from the VMware/NVIDIA partnership announced in October of 2011?  Will it happen with some sort of SR-IOV?  The need for GPU performance is there.  How it will be a achieved, I don’t know.  In the short term, if you need big time GPU power, a physical workstation with a PCoIP host card will work fine.

The performance and wow factor of running a View VM on a tablet is high as well.  If you want to impress anyone, just show this setup on a tablet.  Two or three taps on the tablet and you are in.  But we all know that User Experience (UX) designs for desktop applications were meant for a large screen, mouse, and a keyboard.  It will be interesting to see how the evolution of these technologies continue, so that UX can hit mobile devices in a more appropriate way.  Variations of application virtualization is perhaps the next step.  Again, another exciting unknown.

Also a worthwhile note is competition, not only in classically defined VDI solutions, but access to systems.  A compelling aspect of using View is that it pairs a solution for remote display, and brokering secure remote access into one package.  But other competing solutions do not necessarily have to take that approach.  Microsoft’s “Direct Access” allows for secure RDP sessions to occur without a traditional VPN.  I have not had an opportunity yet to try their Unified Access Gateway (UAG) solution, but it gets rave reviews from those who implement it, and use it.  Remote Desktop Session Host (RDSH) in Windows Server 8 promises big things (if you only use Windows of course).

Among the other challenges is how to implement such technologies in a way that is cost effective.  Up front costs associated with going beyond a pilot phase might be a bit tough to swallow, as technical challenges such as storage I/O deserve attention.  I suspect with the new wave of SSD and SSD hybrid SAN arrays out there, that it might make the technical and financial challenges more palatable.  I wish that I had the opportunity to demonstrate how well these systems would work on an SSD or hybrid array, but the word “pilot” typically means “keep the costs down.”  So no SSD array until we move forward with a larger deployment.

There seems to be a rush by many to take a position on whether VDI is the wave of the future, or a bust that will never happen.  I don’t think its necessary to think that way.  It is what it is; a technology that might benefit you or the business you work for, or it might not.  What I do know is that it is rewarding and fun to plan and deploy innovative solutions that help end users, while addressing classic challenges within IT.  This was one of those projects.

Those who have done these types of implementations will tell you that successful VDI implementations always pay acute attention to the infrastructure, especially storage.  (Reading about failed implementations seems to confirm this).  I believe it.  I was almost happy that my licensing forced me to keep this deployment small, as I could focus on the product rather than some of the implications with storage I/O that would inevitably come up with a larger deployment.  Economies of scale makes VDI intriguing in deployment and operation.  However, it appears to be that scaling is the tricky part. 

What might also need a timely update is Windows licensing.  There is usually little time left in the day to understand the nuances of EULAs in general – especially Windows licensing.  VDI adds an extra twist to this.  A few links at the end of this post will help explain why.

None of these items above discount the power and potential of VDI.  While my deployments were very small, I did get a taste of its ability to consolidate corporate assets back to the data center.  The idea of provisioning, maintaining, and protecting end user systems seems possible again, and in certain environments could have a profound improvement.  It is easy to envision smaller branch office greatly reducing, or eliminating servers at their location.   AD designs simplify.  Assets simplify, as does access control – all with providing a more flexible work environment.  Not a bad combination.

Thanks for reading.

Helpful Links:
Two links on Windows 7 SPLA and VDI

RDSH in Windows Server 2008

VDI has little to do with the Desktop

Scott Lowe’s interesting post on SR-IOV

Improving density of VMs per host with Teradici’s PCoIP Offload card for VMware View

VDI for me. Part 4


Picking up where we left off in VDI for me. Part 3, we are at a point in which the components of View can be installed and configured.  As much as I’d like to walk you through each step, and offer explanations at each point, sticking to abbreviated steps is a better way to help you understand how the pieces of the puzzle fit together.  Besides, others have great posts on installing and configuring the View Connection servers, not to mention the VMware documentation, which is quite good.  The links at the end of the post will give you a good start.  My focus will be to hit on the main areas to configure to get View up and running.

Here is the relationship between the Connection Servers, the clients, and the systems running the agents in my environment.  The overall topology for my View environment can be found in VDI for me. Part 2.

For clients accessing View from the Internal LAN


For clients accessing View from offsite locations


Overview of steps
This is the order I used for deploying the View Components.  To simplify, you may wish to skip steps 3 and 4 until you get everything working on the inside of your network. 

  1. Install Composer on your vCenter Server.
  2. Build a VM and install View Connection Server intended for local LAN access only.
  3. Build a VM and install View Connection Server intended to talk with Security Server.
  4. Build a VM and install View Security Server in your DMZ.
  5. Install View Agent on a VM.
  6. Create a Pool for the VM, and entitle the pool to an user or group in AD.
  7. Connect to the VM using the View Client.

Configuring your first Connection Server (For Internal Access)
From the point that your first connection manager is installed, you may begin the configuration.

  1. Browse out to VMware View Administrator portal on your connection server (https://[yourconnectionserver]/admin) and enter the appropriate credentials.
  2. Drill down into View Configuration > Product Licensing and Usage > Edit License to add your license information.
  3. Register your vCenter Server by going to View Configuration > Servers > Add.  Fill out all of the details, but do not click “Enable View Composer” quite yet.  Click OK to exit.
  4. Go back into Edit the vCenter server configuration, and click “Enable View Composer and Click OK to exit.
  5. In the area where the listing of View Connection servers are listed, select the only View Connection Server on the list, and click “Edit”.  You will want to make sure both check boxes are unchecked, and use internal FQDN and IP addresses only.


Configuring your Second Connection Server (to be paired with Security Server)
During the installation of View on the second server, it will ask what type of Connection Server it will be.  Choose “Replica” from the list, and type in the name of your first Connection Server.

  1. Browse out to the View Administrator Portal, and you will now see a second connection server listed.  Highlight it, and click on Edit.
  2. Unlike the first connection server, this connection server needs to have both checkboxes checked.


Configuring your Security Server (to be paired with your second Connection Server)
Just a few easy steps will take care of your Security Server.

  1. Browse out to the View Administrator portal, highlight the Connection Server you want to pair with the security server, and click More Commands > Specify Security Server Pairing Password.
  2. Install the Connection Server install bits onto your new Security Server.  Choose “Security Server” for the type of Connection Server it will be.  It will then prompt you to enter the internal Connection Server to pair it to.  This is the internal FQDN of the server Connection Server.
  3. Enter the View pairing password established in step 1.  This will make the Security Server show up in the View Administrator Portal.
  4. Go back to the View Administrator portal, highlight the server that is listed under the Security Server, and click Edit.  This is where you will enter in the FQDN desired.  The PCoIP address should be the publicly registered IP address.  In my case, it is the address bound to the external interface of my firewall, but your topology might dictate otherwise.



After it is all done, in the View Administrator portal, you should see one entry for a vCenter server, two entries for the View Connection servers, and one entry for a Security Server.


From this point, it is just a matter of installing the View Agent on the VMs (or physical workstation with a PCoIP host card) you’d like to expose, create a pool, entitle a user or group, and you should be ready to connect.

After you add the VMware View adm templates to Active Directory, a number of tunable settings will be available to you.  The good news in the tuning department is that while PCoIP is highly tunable, I don’t feel it has to be the first thing you need to address after the initial deployment.  With View 5, it works quite well out of the box.  I will defer to this post on some common, View specific GPO settings you might want to adjust, especially in a WAN environment.  The two settings that will probably make the biggest impact are the “Maximum Frame Rate” settings, and the “Build to Lossless” toggle.  I applied these and a few others in order to accommodate our Development Team working on another continent deal with their 280+ms latency. 

The tools available to monitor, test, and debug PCoIP are improving almost by the day, and will be an interesting area to watch.  Take a look at the links for the PCoIP Log Viewer and the PCoIP Configuration utilities at the end of this post.

Tips and Observations
When running View, there is a noticeable increase on the dependence of vCenter, and the databases that support it and View Composer.  This is especially the case in smaller environments where the server running vCenter might be housing the vCenter database, and the database for View Composer.  Chris Wahl’s recent post Protecting the vCenter Database with SQL Log Shipping addresses this, and provides a good way to protect the vCenter databases through log shipping.  If you are a Dell EqualLogic user, it may be helpful to move your SQL DB and Log volumes off to guest attached volumes, and use their ASM/ME application to easily make snaps and replicas of the database.  Regardless of the adjustments that you choose to make, factor this in to your design criteria, especially if the desktops served up by View become critical to your business.

If your connection to a View VM terminates prematurely, don’t worry.  It seems to be a common occurrence during initial deployment that can happen for a number of reasons.  There are a lot of KB articles on how to diagnose them.  One that I ran across that wasn’t documented very much was that the VM may not have enough memory assigned to the video RAM.  The result can be that it works fine using RDP, but disconnects when using PCoIP.  I’ve had some VMs mysteriously reduce themselves back down to a default number that won’t support large or multiple screen resolutions.  Take a quick look the settings of your VM.  Once those initial issues have been resolved, I’ve found everything to work as expected.

In my mad rush to build out the second View environment at our CoLo, everything worked perfectly, except when it came to the View client validating the secured connection. All indicators pointed to SSL, and perhaps how the exported SSL certificate was applied to the VM running the Security Server. I checked, and rechecked everything, burning up a fair amount of time. It turned out it was a silly mistake (aren’t they all?). In C:\Program Files\VMware\VMware View\Server\sslgateway\conf there needs to be a file called This contains information on the exported certificate. Well, when I created the file, Windows was nice enough to append the .txt to it (e.g. The default settings in Windows left that extension hidden, so it didn’t show. By the way, I’ve always hated that default setting for hiding file extensions. It is controlled via GPO at my primary site, but didn’t have that set at the CoLo site.

Next up, I’ll be wrapping up this series with the final impressions of the project.  What worked well, what didn’t.  Perceptions from the users, and from those writing the checks.  Stay tuned.

Helpful Links
VMware View Documentation Portal.  A lot of good information here.

A couple of nice YouTube videos showing a step by step installation of View Composer

How to apply View specific settings for systems via GPO (written for 4.5, but also applies to 5.0)

PCoIP disconnect codes

PCoIP Log Viewer

PCoIP Configuration utility (beta)

More PCoIP tuning suggestions

VDI for me. Part 3


In VDI for me. Part 2, I left off with how VMware View was going to be constructed in my environment.  We are almost at the point of installing and configuring the VMware View components, but before that is addressed, the most prudent step is to ensure that the right type of traffic can communicate across the different isolated network segments.  This post is simply going to focus on the security rules to do such a thing.  For me, access to these segments are managed by a Celestix MSA 5200i, 6 port Firewall running Microsoft ForeFront Threat Management Gateway (TMG) 2010.  While the screen captures are directly from TMG, much of the information here would apply to other security solutions.

Since all of the supporting components of VMware View will need to communicate across network segments anyway, I suggest making accommodations in your firewall before you start building the View components.  Sometimes this is not always practical, but in this case, I found that I only had to make a few adjustments before things were working perfectly with all of the components.

My network design was a fairly straightforward, 4 legged topology. (a pretty picture of this can be seen in Part 2)

Network Leg Contains
External All users connecting to our View environment.
LAN View connection server dedicated for access from the inside. 
View connection server dedicated for communication with the Security Server. 
Systems running the View agent software.
DMZ1 Externally facing View “Security Server”
DMZ4 vSphere Management Network.  vCenter, and SQL databases providing services for vCenter, and View Composer.

For those who have their vSphere Management network on a separate network by way of a simple VLAN, your rules will be simpler than mine.  For clarity, I will just show the rules that are used for getting VMware View to work.

Before you get started, make sure you have planned out all of the system names and IP addresses of the various Connection Servers, VM’s running the View Agent.  It will make the work later on easier. 

Creating Custom Protocols for VMware View in TMG 2010
In order to build the rules properly, you will first need to define some “User-Defined” protocols.  For the sake of keeping track of all of the user defined protocols, I always included the name “View” (to remember it’s purpose), the direction, type, and the port number.  Here was the list (as I named them) that was used as a part of my rule sets.

VMware View Inbound TCP&UDP (4172)
VMware View Outbound (32111)
VMware View Outbound (4001)
VMware View Outbound (8009)
VMware View Outbound (9427)
VMware ViewComposer Inbound (18443)
VMware ViewComposer Outbound (18443)
VMware ViewPCoIP Outbound (4172)
VMware ViewPCoIP SendReceiveUDP (4172)

Page 19 of the VMware View Security Reference will detail the ports and access needed.  I appreciate the detail, and it is all technically correct, but it can be a little confusing.  Hopefully, what I provide will help bridge the gap on anything confusing in the manual.  My implementation at this time does not include a View Transfer Server, so if your deployment includes this, please refer to the installation guide.

Creating Access Rules for VMware View in TMG 2010
The next step will be to build some access rules.  Access rules are typically defining access in a From/To arrangement.  Here are what my rules looked like for a successful implementation of VMware View in TMG 2010.


Creating Publishing rules for VMware View in TMG 2010
In the screen above, near the bottom, you see two Publishing rules.  These are for the purposes of securely exposing a server that you want visible to the outside world.  In this case, that would be the View Security Server.  The server will still have its private address as it resides in the DMZ, but would take on one of the assigned public IP addresses bound to the external interface of the TMG appliance.  To make View work, you will need two publishing rules.  One for HTTPS, and the other for PCoIP.  A View session with the display setting of RDP will use only the HTTPS publisher.  A View session with the display setting of PCoIP will use both of the publishing rules.  Page 65 of the View 5 Architecture Planning Guide illustrates this pretty well.

In the PCoIP publishing rule, notice how you need both TCP and UDP, and of course, the correct direction.


My friend Richard Hicks had some great information on his ForeFront TMG blog that was pertinent to this project. ForeFront TMG 2010 Protocol Direction Explained is a good reminder of what you will need to know when defining custom protocols, and the rule sets that use them.  The other was the nuances of using RDP with the “Web Site Publishing Rule” generator.  Let me explain.

TMG has a “Web Site Publishing Rule” generator that allows for a convenient way of exposing HTTP and HTTPS related traffic to the intended target. This publisher’s role is to protect by inspection. It terminates the session, decrypts, inspects, then repackages for delivery onto its destination. This is great for many protocols inside of SSL such as HTTP, but protocols like RDP inside SSL do not like it. This is what I was running into during deployment. View connections using PCoIP worked fine. View connections using RDP did not. Rich was able to help me better understand what the problem was, and how to work around it. The fix was simply to create a “Non-Web Server Protocol Publishing Rule” instead, choosing HTTPS as the protocol type.  For all of you TMG users out there, this is the reason why I haven’t described how to create a “Web Listener” to be used with a traditional “Web Site Publishing Rule.”  There is no need for one.

A few tips in with implementing you’re your new firewall rules.  Again, most of these apply to any Firewall you choose to use.

1.  Even if you have the intent of granular lockdown (as you should), it may be easiest to initially define the rule sets a little broader.  Use things like entire network segments instead of individually assigned machine objects  You can tighten the screws down later (remember to do so), and it is easier to diagnose issues.

2.  Watch those firewall logs.  Its easy to mess something up along the way, and your real time firewall logs will be your best friend.  But be careful not to get too fancy with the filtering.  You may be missing some denied traffic that doesn’t necessarily match up with your filter.

3.  You will probably need to create custom protocols.  Name them in such a way that they are clear that they are an incoming or outgoing protocol, and perhaps whether they are TCP, or UDP.  Otherwise, it can get a little confusing when it comes to direction of traffic.  Rule sets have a direction, as do the protocols that are contained in them.

4.  Stay disciplined to rule set taxonomy.  You will need to understand what the rule is trying to do.  Consistency is key.  You may find it more helpful to name the computer objects  the role that they are playing, rather than their actual server name.  It helps with understanding the flow of the rules.

5.  Add some identifier to your rules defined for View.  That way, when you are troubleshooting, you can enter “View” in the search function, and it quickly shows you only the rule sets you need to deal with.

6.  Stick the the best practices when it comes to placement of the View rules into your overall rule sets.  TMG processes the rules by order, so there is some methods to make the processing most efficient.  They remain unchanged from it’s predecessor, ISA 2006.  Here is a good article on better understanding the order.

7.  TMG 2010 has a nice feature of grouping rules.  This gives the ability of a set of contiguous rules to be seen as one logical unit.  You might find this helpful in most of your View based rule creation.  I would probably recommend having your access rules for View in a different group than your publishing rules.  This is so that you can maintain best practices on placement/priority of rule types.

8.  When you get to the point diagnosing what appear to be connection problems between clients, agents, and connection servers, give VMware a call.  They have a few tools that will help in your efforts.  Unfortunately, I can’t provide any more information about the tools at this time, but I can say that for the purposes of diagnosing connectivity issues, they are really nice.

I also stumbled upon an interesting (and apparently little known) issue when you have a system with multiple NICs that is also running the View agent.  For me, this issue arose on the high powered physical workstation with PCoIP host card, using View as the connection broker.  This system had two additional NICs that connected to the iSCSI SAN.  The PCoIP based connections worked, but RDP sessions through View failed, even when standard RDP worked without issue.  Shut off the other NICs, and everything worked fine.  VMware KB article 1026498 addresses this.  The fix is simply adding a registry entry

On the host with the PCoIP card, open regedit and add the following entry:

Add the REG_SZ value:

If you experience issues connecting to one system running the View Agent, but not the other, a common practice is to remove and reinstall the View Agent.  Any time the VMware Tools on the VM are updated, you will also need to reinstall the agent. 

More experimentation and feedback
As promised in part 1 of this series, I wanted to keep you posted of feedback that I was getting from end users, and observations I had along the way. 

The group of users allowed to connect continued to be impressed to the point that using it was a part of their workday.  I found myself not being able to experiment quite the way I had planned, because users were depending on the service almost immediately.  So much for that idea of it being a pilot project.

The experimentation with serving up a physical system with PCoIP using VMware as a connection broker has continued to be an interesting one. There are pretty significant market segments that demands high powered GPU processing. CAD/CAE, visualization, animation, graphic design, etc have all historically relied on client side GPUs. So it is a provocative thought to be able to serve up high powered graphics workstations without it sitting under a desk somewhere.  The elegance of this arrangement is that once a physical system has a PCoIP host card in it, and the View Agent installed, it is really no different than the lower powered VM’s served up by the cluster.  Access is the same, and so is the experience.  Just a heck of a lot more power.  Since it is all host based rendering, you can make the remote system as powerful as your budget allows.  Get ready for anyone who accesses a high powered workstation like this to be spoiled easily.  Before you know it, they will ask if they can have 48GB of RAM on their VM as well. 

Running View from any client (Windows Workstation, Mac, Tablets, Ubuntu, and a Wyse P20 zero client) proved to give basically the same experience.  It was easy for the end users to connect.  Since I have a dual name space (“” from the outside, and “view.local.lan” from the inside), the biggest confusion has been for laptop users remembering which address to use.  That, and reminding them to not use the VPN to connect.  A few firewall rules blocking access will help guide them. 

One of my late experiments came after I met all of my other acceptance criteria for the project.  I wanted to see how VMware View worked with linked clones.  Setting up linked clones was pretty easy.  However, I didn’t realize until late in the project that a linked clone arrangement of View really requires you to run a Microsoft KMS licensing server.  Otherwise, your trusty MAK license keys might be fully depleted in no time.  There is a VMware KB Article describing a possible workaround, but it also warns you of the risks.  Accommodating for KMS licensing is not a difficult matter to address (except for extremely small organizations who don’t qualify for KMS licensing), but it was something I didn’t anticipate.

I had the chance to do this entire design and implementation not once, but twice.  No, it wasn’t because everything blew up and I had no backups.  My intention was to build a pilot out at my Primary facility first, then build the same arrangement (as much as possible) at the CoLocation facility.  What made this so fast and easy?  As I did my deployment at my Primary facility, I did all of my step by step design documentation in Microsoft OneNote; my favorite non-technical application.  Step by step deployment of the systems, issues, and other oddball events were all documented the first time around.  It made the second go really quick and easy.  Whether it be my Firewall configuration, or the configurations of the various Connection Servers, the time spent documenting paid off quickly.

Next up, I’ll be going over some basic configuration settings of your connection servers, and maybe a little tuning.

VDI for me. Part 2


In VDI for me.  Part 1, I described a little bit about our particular interests and use cases for VDI, and how we wanted to deploy a pilot project to learn more about it.  So now what?  Where does a person start?  Well, deployment guides and white papers for VMware View will tell you all of the possibilities.  They are thorough, but perhaps a bit tough to decipher when trying to understand what a simple VDI arrangement might look like.  So I’m just going to focus on how I built out a simple arrangement of VMware View, and how the components of VDI listed in my first post fit together.

Topology of VMware View
The Topology of View should fit in very nicely with what what you already have for a virtualized infrastructure.  If you are already running a vSphere cluster in your environment, its really just a matter of building your supporting components, which in a very simple or small deployment might be an instance of Composer, and a View Connection server or two.


VMware deployment guides will have all of the details of these components, so I will try to keep it brief. 

Connection Server.  This is the VM that is the broker that your View Client software will connect to, and will be available to users inside your network to connect to (e.g. “view.corp.lan”).  This is a domain joined system, and should probably reside on your primary LAN network. 

Security Server.  This is another VM that is running the Connection Server component but in a slightly different mode.  It’s purpose is to act as the secure proxy for external clients to connect to your internal systems presented by View.  It should only be used for clients outside of your network to connect to (e.g. “”).  This server should not be joined to the domain, and should reside in a secure DMZ segment of your network that you have full granular control of ingress and egress traffic.

Composer.  This is the server that is behind the magic of using a minimal storage footprint to issue out many Virtual Desktops – by way of linked clones.  For my simple deployment, I chose to install it on my VM running vCenter.  I also chose to stick to the basics when it came to configuring Composer, as I really wanted to focus on the behavior of the View Client itself, not experiment how best to deploy hundreds of desktops with linked clones.  Composer needs a SQL Server, so you can use the database server that serves up vCenter, or something else if you wish.

Transfer Server.  This is a dedicated server that is used for “offline mode” laptop users or “Bring your Own PC” (BYoPC) arrangements.  Its optional, but gives the ability for a laptop user who normally uses a VDI VM that traditionally lives on the infrastructure, to be “checked out” to the laptop.  It uses all of the local resources of the laptop, and when/if it has the ability to phone home via an internet connection, it can send and update of the VM.  It is an optional feature, but a compelling one for those who might ask, “but what if I need my Virtual Desktop, and I have no internet connection?” or if you are considering a BYOPC arrangement.  It does not have to be joined to the domain, and its placement depends on the use case.

View Agent.  This is the tiny bit of software installed on a system that you want VMware View to present to a user.  In its simplest form, it can just be loaded onto an existing VM, or something served up from Composer.  If you need really high end horsepower with perhaps intensive graphics, it can be installed on a physical workstation with a PCoIP host card, so that VMware View can serve it up just like any other resource (more on this in a later post).  Right now, the agent software can only be installed on a Windows based system. 

View Client.  This is the application used by the end user to connect to their Virtual Desktop.  Currently the official client software is limited to Windows based machines, and iPads.  However, there is a Technology Preview edition for Mac OSX, and some distributions of Linux.  Note that if you are using a PCoIP Zero client, such as a Wyse P20, there is no software that you have to install on it.  Just connect it to the network, and connect to the internal View Connection Server, and you are good to go.

Interestingly enough, If you have a need to access your VDI systems from outside of your LAN, (thus requiring you to run the Security Server in a DMZ), you will need a separate/additional VM (domain joined) on the inside of your LAN that is running the Connection Server software, and is used exclusively by the Security Server.  The reason it cannot use the Connection Server above that was designated for internal use is that there are toggles that must be set based on the type of connection that is going to occur.  These toggles are mutually exclusive. (I’ll address this more on later posts)  You can try to use just one Connection Server coupled with a Security Server for both internal and external access, but traffic will not flow efficiently, and your performance will be affected by this. 

I think VMware might need to review their architecture of the Connection Servers.  Having three VM’s (which doesn’t even include the View Transfer Server) serve up internal and external access probably doesn’t seem like much when you have hundreds of clients, but for a pilot or a smaller deployment, it is overkill, and could be reduced to at least two with some tweaks in the Connection Server software.  Perhaps it wouldn’t be that big of a deal if they were just VM appliances.  (hint hint…)

The Software
I chose the “View Premier Starter Pack” which included all of the things needed to get started. View Composer, ThinApp, View Persona Management, vShield Endpoint, and “View Client for Local Mode”  Also included is a license of vSphere Enterprise Plus for Desktop Standalone licenses, and vCenter (special licensing considerations apply).  The result is that you get to try out VDI on a single vSphere host, with up to 10 VDI clients, and all of the tools necessary to experiment with the features of View.  From there, you would just purchase additional client licenses or vSphere host licenses as necessary.  It is really a great bundle, and an easy way to begin a pilot project with everything you need.

Placement of systems
More than likely, you will be presenting these systems to the outside world for some of your staff.  This is going to involve some non-VMware related decisions and configurations to get things running correctly.  Modifications will need to be made with your Firewall, DNS, SQL, vSphere, and AD.  If you are a smaller organization, the Administrator may be the same person (you).  Daunting, but at least you cut out the middle man.  That’s the way I like to think of it.  Nevertheless, you will want to at least exercise some care in the placement of systems to minimize performance issues, making sure traffic is running efficiently, in a secure manner.

There is going to be quite a bit of communication between the Connection servers, vCenter, the systems running the agent, and the clients running the View Client.  VMware publishes a nice document on all of the ports needed for each type of connection.  It’s all technically there, but does take a little time to decipher.  I stared at this document endlessly, while keeping a close I on my real-time firewall logs as I attempted to make various connections.  I really think I needed the two in order to get it working. 

As mentioned earlier, you should really have the Security Manager VM sitting off in a protected network segment off of your firewall.  For me, my primary edge security device is a Microsoft ForeFront Threat Management Gateway (TMG) Firewall software running on a nicely packaged appliance by Celestix Networks.  I’ve said it before, and I’ll say it again.  It is such a shame that Microsoft does not get its due credit on this.  My security friends in high places have consistently stated that there is simply nothing better when it comes to robust, flexible protection across the protocol and application stacks than TMG.  I feel fortunate to be working with the good product that it is.  The information that I share throughout this series will be based on use of TMG, but the overall principals will be similar to other edge security solutions.

How much you need to modify your firewall depends on how you place some of these systems.  Since these View related components will be interacting with vCenter, you will have to plan accordingly.  Some vSphere environments simply have their vSphere Management Network (aka Service Console network) simply behind a VLAN and routed directly to the LAN, while others have it on a separate network leg only accessible by a physical router or firewall.  I fall in the latter category, as my old switches never had the capability to do interVLAN routing.  Happily, these old switches have since been replaced, but I my vSphere Management network stills resides in an isolated network leg, and my Firewall settings for VMware View will accommodate for that.

Early impressions
While I wanted to deploy the systems correctly, and minimize floundering, I also wanted to see some early results.  After I built up the systems to Manage the environment, and worked out the connectivity issues, I thought I’d go ahead and install the View Agent on a few of the existing VMs used by our Software Developers located on another continent.  These Developers have played a very important role to the companies development efforts, so anything I could do to improve their user experience was a plus.  Previously, their work environment consisted of their own laptop connected to our network via a PPTP based VPN, then using RDP to connect up to a VM provisioned for them.  Their VM has all of their Development applications and tools, which allows for the data to be kept in the datacenter.  Oh, and their typical latency?  Around 280ms on average.  Yeah, their connection is that terrible. 

So my first tests really involved nothing more than installing the VMware View agent on their VM, then giving them the web address to download the VMware View Client, and a few steps on how to connect.  I also reminded them not to log into the VPN, as that was no longer necessary.  I asked them to try both RDP and PCoIP over their connection, along with some anecdotal feedback whenever they had a chance. 

  • Within 20 minutes they responded saying things like, “At first glance it works awesome” and “Much much faster than VPN + RDP.
  • That same day they told the rest of their team members about it, and during a Development review, stated, “please don’t take it away!
  • Further feedback included comments like “This feels at least 10x faster than the older method” along with “I wouldn’t dream of playing a video on the old RDP, and rotating a 3D view in our software would’ve made you want to bang your head against a wall.  This [new] way is great.
  • Their impressions over their highly latent connection was that using the View client with PCoIP was much more responsive than the View client with RDP.

What is really interesting was that I didn’t make a single modification to tune PCoIP, nor did I adjust their VMs in any way.  That is the power of a new approach to remote display rendering.  I do not know the exact thresholds that Teradici (the makers of the PCoIP protocol) were shooting for when it came to latency, but I’m willing to bet that nearly 300ms of latency was outside of their typical use case.  It is quite the compliment to the power of PCoIP.  Needless to say, once it was tested, I couldn’t bear to take this away from the Remote Developers, so I did my best to work my deployment around their work days.  This was really exciting because these were just normal VM’s that I was serving up.  I didn’t have a chance at this point to serve up the high end workstation with the VMware View Agent installed on it – something I was really looking forward to.

Some more lessons and observations
I’ve witnessed what others have experienced.  It is far too easy to let a pilot project turn into a production environment.  FAST.  With reactions like I shared above, it happens right before your eyes.  The neat features that were demonstrated were must-haves almost overnight.  No wonder so many small VDI deployments ultimately suffer in some form or another as they grow.  So if you are contemplating even a pilot deployment, keep this in mind.  I’m doing my best to contain expectations until I have one of the key ingredients in place; fast storage targeted for IOPS hungry VDI deployments.  For me this means a Dell EqualLogic PS6100XV hybrid array.  Until that time, I am not going to even consider expanding this beyond the 10 instances that comes with the View Premier Bundle.

It was nice to see that during my early deployment, that VMware released a “Technology Preview” edition of the View 5.0 Client for Mac OSX, as well as a version for Ubuntu.  It is a nice step forward, but it doesn’t go far enough.  A fully supported View Client needs to be provided for Redhat clones, and debian based linux distributions.  Likewise, the View Agent (the bit of software installed on the VM or the physical workstation that makes the system available via View) is limited to only Windows Operating Systems.  It is my position that having an agent for Linux distributions would serve VMware quite well – much better than they even know.  I’m hoping this is in their feature backlog.

Next up, I’ll be going over some typical View Administrator configuration settings, Firewall settings to get View and PCoIP flowing correctly, as well as my experimentation with a physical workstation with a PCoIP host card and a VMware View Agent installed so that it can be brokered by VMware View.


Helpful links
A recently released post by VMware describing exactly the benefit with hardware based systems that we are trying to exploit.

VMware View 5.0 Documentation Portal

A comprehensive guide to ports needed to be opened with View 5 and it’s related services

The readme for the View Client Tech Preview for Mac OSX

EqualLogic hybrid arrays in VDI testing

VDI for me. Part 1


It doesn’t take long for those in IT to see the letters “VDI” show up on the radar.  Maybe a morning digest of RSS or twitter feeds might be all it takes.  But the subject of Desktop Virtualization can be a difficult puzzle to piece together if you haven’t already invested in it.  There is a lot to sort through, even if you are contemplating just a small pilot project.  VDI deployments can be a highly visible IT project (in user expectations, experiences, and financial commitment), so if you are contemplating a pilot project, you’ll want to stack the odds of success in your favor, and make that first impression a good one.  As I embarked on my own VDI project, I noticed that while there was much to be found on the net regarding VDI in general, one thing seemed remarkably absent; information on how pilot projects are approached, and actually deployed.  My posts won’t be dissecting any CAPEX or OPEX numbers, nor will it attempt to prove that VDI is the perfect solution for every scenario.  But rather, I want to give you a glimpse to what a VDI pilot project actually looks like. 

As I mentioned on previous posts, my recent efforts to upgrade my hypervisor to vSphere 5 was for one real purpose.  I wanted to deploy a Proof of Concept (PoC) for VDI in our environment using VMware View 5.  The company I work for makes industry leading data visualization and simulation analytics software.  Our customers, who are often scientists and engineers, need to understand simulation results to make smart design decisions.  The software is typically installed on a user’s local workstation (Windows, Linux, Mac, or Unix) and interacts with data that is either on the network, or if performance dictates, their local system.  The nature of the problems that are trying to be solved, along with our software, can demand high computational horsepower with high-end graphics.  Many in the CAE/CAD industries are familiar with the demands of the local workstation performing the work.  But there are some scenarios with VDI that make it quite compelling that are not brought up very often.  We wanted to understand those benefits better.

Much of the interest internally transpired from a technology lunch-and-learn session I gave to our staff.  My motives for doing such a thing were two-fold.  1.)  I wanted everyone interested (not just key stakeholders, but software developers, technical support, etc.) to be aware of some things I knew they weren’t aware of, and 2.), I thought it would be fun.  I covered anything from how virtualization has changed the Datacenter, to demonstrating how common assumptions about our own internal systems, as well as our customers, may be highly misguided based off of new technologies.  While the focus was the Datacenter, I translated much of that into how these changed might affect end users.  It was a tremendous success, and even required an encore presentation for those who missed it.  This piqued the interest of the key stakeholders, as they understood it could have a significant impact on our product roadmap.  Combine this with the impressive “technology previews” of things like “AppBlast” and “Octopus” at the 2011 VMworld, as well as recent partnerships announced by VMware and NVIDIA for providing hardware based GPU acceleration to hypervisors, and we had something we needed to look into.

Due to the type of software we make, we had some unique use cases that we not only wanted to understand better ourselves, but perhaps provide our customers with solutions on a better way to do things.  I had several objectives I wanted to achieve.

  • Understand how we could improve on our own internal efficiencies and toolset to our staff.  We have taken a fairly aggressive approach already to provide ubiquitous access to everyone in our office (tablets for everyone, and their use encouraged in meetings, etc.).  We also have a great team of software developers on another continent, and are always looking for new ways to integrate them with our systems.  We needed to look at tools that not only help that, but help mitigate the the operational expenses and security challenges of traditional workstations to end users.  These seem to be the mostly common reasons cited when organizations are looking into VDI.
  • Demonstrate to our customers possibly “a better way” to work.  Believe it or not, many consumers of applications that have large data still either download the data locally, then work on it, or rely on local high speed networks to transmit the data.  What happens if the data, and the system used to present the data lived in the Datacenter?  These days, under a virtualized infrastructure, network communications may never even see an Ethernet cable or a physical switch, and the results can be impressive.  This is the benefit that most of us have enjoyed while virtualizing our server infrastructure.  So if the new type of client server arrangement only had to deal with transmitting screen data, then a world of possibilities opens up.
  • Demonstrate our software.  We rely heavily on feedback from new and existing customers on our products, and we need them to try our software.  If your customer base runs on extremely sensitive or secure networks, you’ll find that a pattern emerges.  They won’t let you install trial or pre-release software on any of their machines.  We’ve resorted to some pretty crude methods to help them test our software (shipping some old laptops), but obviously, this doesn’t scale well, and frankly, doesn’t give a good impression for a company who strives for innovation.  
  • Understand new display protocols, and how this could impact how datacenters are constructed.  Remote displays are a big concern in the visualization, CAE, and CAD worlds.  Years of dealing with various display protocols have resulted in similar experiences.  Most share the trait of using connection oriented protocols to transmit rendering instructions.  They tend to suffer on connections beyond the LAN that may have high latency and packet loss.  There are new approaches to help change this.

GPUs, Remote Displays, and Client Server
Most heavy hitting graphics software packages have historically relied HEAVILY on GPU power provided by a real video card.  This has been under the assumption that they’d always be there, and that GPUs are incredibly powerful.  Well, as you can imagine, that doesn’t play to the strengths of traditional virtualization, where video has barely been an afterthought (at least so far).  Not a big deal when you were virtualizing your server infrastructure, but a pretty big deal when you are trying to virtualized desktops.  Task workers may never notice the difference, but users of high end graphics will.

Remote display protocols come in a variety of forms.  Some are definitely better than others, but most suffer from fundamental issues of being based on TCP.  The rules of TCP guarantee reliable and orderly delivery of packets to be reassembled at the target.  Network communication wouldn’t have gotten very far if it didn’t have a way to guarantee delivery, and while this is great for most things, it doesn’t shine when it comes to video, animations, or high speed screen refreshes.

So why would a software company be so curious about this new set of technologies? Technologies that for the most part, we have no control of?  The answer is that it can change how one architects software. For example:

  • Sometimes there are calls for a true two-piece, client server software, where the client application and the server application are working together to perform their own dedicated tasks, but in different locations.  The goal with this type of client server arrangement was to use recourses as efficiently as possible. It is not as common as it once was, and history has demonstrated that this approach can be complex, wrought with problems, and very expensive to develop and maintain, especially when dealing with heterogeneous environments.  It may work in some environments, but not in others.  It may or may not use chatty protocols that simply don’t work well over poor network conditions. The further the two are separated, the more you introduce issues.  Firewalls, network traversal, etc. all complicate matters.
  • The other form of client server that people are most familiar with is where the application and computing power is in one location (e.g. The desktop) while the files being accessed are on a file server.  While it is very clear what is doing the work, it begins to falter with large data, or traversing geographic boundaries.  Look at the traffic generated by opening up a modest sized spreadsheet over a VPN connection, and you’ll see what I’m talking about.

With the trends on virtualizing infrastructures, the datacenters have collapsed to be more centrally located. If the client system (a workstation, or in this case, a VM) is now in the same virtual space as where the data lives (the server, or in this case, a VM), it can have an impact on application design.  Instead of using “client rendering” the model uses “on-host” rendering, where all of the heavy lifting is performed inside the Datacenter. Meanwhile, the endpoint devices, such as laptops, tablets, zero or thin clients, which do not need to have much horsepower, are really just acting in a presentation type role. It’s a bit like wondering how much CPU power your TV has.  Answer… It doesn’t matter, as the score of the baseball game is being rendered on-host.

The Plan
I wanted to make sure that I wasn’t claiming that VDI will change everything, and that every single person will be working off of a zero client in 12 months.  Nor was I suggesting that running a virtual desktop on a tablet is the ideal interface for interacting with a desktop.  That wasn’t my point.  It is ultimately how applications are delivered to the end user.  I wanted to help everyone understand that for the majority of our customers, they simply work on the systems that are provided to them by their IT Department.  Rarely does an end user get line-item-veto power on what they use for a computer, and since we know IT infrastructures are changing, the evidence suggests that what the customers will be using is going to change as well.  In fact, there is pretty good likelihood that we will have a customer who shows up one day to their office with nothing more than a zero client in front of them.

The idea behind the PoC was to invest a minimum amount to provide as much information possible to make smart decisions in the future.  In that spirit, I also plan on revealing results on these posts along the way as well.  I knew my initial project wasn’t going to include the time to look into every feature of a fully deployed VDI.  I was going to stick to the basics, and see how they work.

  • Test access to VM’s using VMware View as the connection broker (via PCoIP, and RDP)
  • Test access to high end physical workstations with VMware View as the connection broker. (via PCoIP, and RDP)
  • Test these systems from a variety of endpoints.  From existing PC’s, to zero clients, to Linux desktops and Mac clients, to wireless tablet devices and smart phones.
  • Test these systems from a variety of connection scenarios.  A connection to something on the LAN is very different than something on another continent. 

Bullet number 2 may have caught your eye.  Creating a unified remote display experience isn’t limited to just virtual machines.  With the power of PCoIP, and using VMware View as a connection broker, one can house high end workstations IN the Datacenter.  They would be a 1:1 assignment (1 active user to 1 workstation) like a traditional workstation, but they’d be close to the storage (perhaps even directly connected to the SAN if desired), and offer the full GPU capabilities of the workstation.  Access to them would be no different than if they were VMs.  In fact, the end user may never know if it is physical, or virtual.  It’s a provocative thought for users of CAD, solid modeling, or simulation analytics software.

Components of my Pilot Project

  • VDI software.  VMware View 5 Premier (their 10 client “starter pack”) running on vSphere 5.0 infrastructure.  This would offer the abilities of the connection broker, the variety of connection protocols (PCoIP, and RDP) among other features.
  • Back-end storage.  For our small pilot project, this was limited to our existing Dell EqualLogic SAN arrays.  They would be fine for this small pilot.  However, I have no illusions about the storage I/O requirements of VDI at a larger scale, and hope that if things go well, a super-fast EqualLogic hybrid SSD/SAS array is in our future.  More on the subject matter of storage and it’s importance to the success of VDI in future posts.
  • Firewalls.  Not to be overlooked, this plays a significant role in how you can present, and secure content.  I have the good fortune to be working with what I believe to be the best of breed.  Microsoft Forefront Threat Management Gateway (TMG) 2010 running on a Celestix MSA appliance.  (more on this in future posts)
  • HTML5 presenter.  While View 5 doesn’t natively support HTML5, I wanted to see what this was like.  Not only would give aid the ability to evaluate our software, but give our developers some insight on how HTML5 may play a part in virtualized, application delivery. (e.g. AppBlast.  Gee, can you tell I’m antsy to get me hands on this?) For this experiment, I’ll be using Ericom’s Access Now for VMware View.
  • Zero Client.  For this, I will be using a Wyse P20.
  • PCoIP host card.  This is an eVGA HD02 PCoIP host card installed on a higher end workstation I will be using to performance against a high end workstation sitting remotely in the Datacenter, and brokered by VMware View.
  • My Primary Site, and my Colocation site.  The CoLo site is not used for anything other than where my offsite SAN replicas go to. My plan is to change that.  Long term intentions are to house services that are more appropriate for that location, while in the near term, I will be housing part of my VDI pilot there.

Early lessons learned
If you are reading this post, more than likely you are well entrenched in the world of virtualization.  Never underestimate that the lack of knowledge in this arena by your coworkers or stakeholders.  About 3 years ago, I started giving a monthly IT review to everyone in our company on what is going on in IT.  This helps dispel the myths behind the giant IT curtain, and possible gives some insight as to the complexities of modern environments.  But no matter how much information I provide, I’m constantly challenged by these technologies, with the occasional question of, “now who is VMware, and what do they do?”  or “What’s a SAN?"  Be prepared to repeat your message several times over.

I would also emphasize VDI as not being an either/or scenario.  It is another form of a computing environment that provides unique capabilities to deliver applications and content.  We know that there are many vehicles for this already, and it continues to evolve.  So in other words, no need to make bold claims about VDI.  This also keeps you out of the business of predicting the future – not a favorable occupation in my book. 

As you have the opportunity to have users try out various use cases, you may have to throttle any over exuberance on the user’s behalf.  Large deployments are different than pilots.  The end user may see the brilliance of the solution, while the budget line owners see nothing but the large capital investment. 

Coming up
In upcoming posts, I’ll share how I chose to design a pilot VDI arrangement for our testing internally, externally, and how we plan to use it for our own internal needs, as well as our customers.  I have no idea how many parts this series is going to be, so bear with me.  What I hope to do is to give you a better understanding of a VDI pilot project in the real world, providing enough detail to be helpful with your own planning.

The VMware, NVIDIA Partnership announcement

Planning for VDI has little to do with the Desktop 

VMware vSphere 4.1 Networking Performance