Translation of On App-V Performance

Last week, I wrote a blog post On App-V 5 Performance which received a lot of attention. From the response, it is pretty clear that my message was not understood. I am a geek, so it is not that unusual. So let’s try to fix that here. Here is an English translation of my thoughts:

The message that I attempted to convey, was part of an ongoing conversation that I have been having with customers in regard to whether to use App-V 4.6 or 5.0. This has been an evolving conversation over time, and I like to separate my views on the question depending on the customer. For the most part, it is a question of whether the customer is using App-V 4.6 and thinking of Migrating to 5.0, or if this is the first implementation of App-V. The message was intended for those thinking of upgrading.

App-V 5, released in November of 2012, was a complete rewrite of the product. Virtualizing applications has always been difficult; while it solves sooooooo many problems it is an unnatural act that introduces its own set of challenges. Microsoft App-V does this better than anyone else, but after 12 years it was time for a fresh start at it. Being a bold rewrite of the app virtualization stack, companies tend to think of 5.0 as more of a 1.0 like product (it really is not, but perception is a funny thing). Microsoft understood this and released a new version of 4.6 (4.6 SP2) to allow customers a choice.

Initially (late 2012), I was quite cautious with customers. For those on 4.6 I recommended playing with 5.0 in the lab to get used to it, but to wait until we had some fixes in a service pack or new release before going to production. For companies new to App-V at that point it was a bit of a tossup: Live with the known issues of 4.6 or the known and unknown issues in 5.0, followed by a conversion later on when you do move to 5.0.

By January of this year, we started to get more comfortable and my advice for new customers changed to a suggestion that they start with 5.0 now. I didn’t change my advice to existing customers at that time, other than to suggest that their willingness to be cutting edge should be considered.

When Service Pack 1 came out in the spring, my message was simply that this wasn’t the release we were waiting for to fix things. SP1 primarily added additional language packs and office support but for the most part did not address the issues we were concerned about.

The Beta for SP2 contains changes for many of the issues that concerned us with 5.0. My messaging to companies then changed. I started saying that they (those customers on 4.6) should probably start getting ready to make the switch once SP2 comes out. (For the record: we don’t know when that will be but it is easy to guess that it has to come out in the same timeframe as Windows 8.1 this fall). Of course that advice depends on what else you have on your plate at the time.

Meanwhile, existing 4.6 customers that made the move using SP1 started making noise about performance issues in certain situations. While we knew that there were publishing performance issues, I thought that these were perceived performance issues in that the “stutter start menu” took a lot longer. The simple kinds of app testing I do in my lab actually seem to perform better than with 4.6, and some simple performance tests that I had run showed a reduced appvirt overhead in a number of important categories.

I had been hearing a lot of questions from companies, anticipating the move to 5.0 this fall, about how to deploy. 5.0 has options of different servers, new package optimization features, and the Shared Content Store mode to play with, and the best way to deploy is a complex question.

I had been suggesting that Shared Content Store was probably a great choice for any App-V client running inside the data center, as in XenApp (RDS), or VDI clients. But as far as I know, nobody had done any real performance testing on it. This was the genesis for the research and book.

What I found, in some isolated and unusual situations, is that the Beta version has some performance issues. I believe (but have not validated) that these are also present in the current release and are related to (if not the cause of) the publishing performance issues. These have been raised with Microsoft, and we should hope to see them resolved, if not in SP2 then sometime after that.

Until we see the fixes and can test them, I wanted to say that I really cannot state whether Shared Cache Mode, or Pre-caching should be used in any given scenario. Or if VMs should have one or many CPUs. So let me make two clearer statements this time:

  • I still think that new customers should go to App-V 5.
  • For existing 4.6 customers, my message is (and was intended to be in that blog post) that you want to wait for SP2, but you will need to test the final SP2 release for performance in the scenarios you intend to deploy.

By Tim Mangan

Tim is a Microsoft MVP, and a Citrix CTP Fellow. He is an expert in App-V and MSIX.