Uses cases server 2008r2

FAQ, getting help, user experience about FancyCache
Post Reply
Sven
Level 1
Level 1
Posts: 2
Joined: Mon Mar 25, 2013 3:17 pm

Uses cases server 2008r2

Post by Sven »

We are currently looking into ways of increase the performance of some XenApp6.5/AppV-5 Windows 2008r2 Servers.

These servers are running on HP blades Xeon E5520 Dual Socket Quad Core 2.26 Ghz with 72Gb Ram and two disks per blade (72gb 15k SAS6g).
With 16 threats and 72gb of RAM those disks are definitely the slowest component. And even SSD's will have some disadvantages over these disks.

I searched the forum and only found use cases of Windows client systems and gaming/apps etc...

We will start testing the FC 0.8 beta and are curious if anyone already tested FC in a 2008r2 server or Citrix XenApp/Terminal Server environment? The way AppV-5 works could benefit extremely from read-cache imo. And the write back would overcome the 2 disk I/O (disk queue) bottle neck.
dustyny
Level 8
Level 8
Posts: 118
Joined: Sun Sep 02, 2012 12:54 am

Re: Uses cases server 2008r2

Post by dustyny »

I'm a big fan of the product but it's not ready for production. I'm using it for pre-production testing, I tend to run about 8vms on average but I've spun up 50 VMs and it could have handled a lot more. I haven't done any real load testing so I can say how it would react if you put users on it. It's pretty amazing product but I've been using the current beta for nearly a year now..
3 Win2012 HyperV hosts, 64GB RAM,
1 file server (SMB3), 8x1TB HDDs RAID 5 and 8x120GB SSDs JBOD, across two SAS storage controllers (no battery backup), 32GB RAM.
Storage traffic goes over a 40Gbs Infiniband, network using RDMA

Here are some of my benchmarks showing different config results.
viewtopic.php?t=1278

I'd also advise against the write defer, it may solve your problem but it'll put your data at risk. You're much better off with a good storage controller with a bunch of ram and a battery backup module.

What's your concern with SSDs? I've been using them in production for years and I'd go all SSD if I could.
Sven
Level 1
Level 1
Posts: 2
Joined: Mon Mar 25, 2013 3:17 pm

Re: Uses cases server 2008r2

Post by Sven »

Thanks for your post, I will look in to your numbers.

We're willing to test :) and who knows when performance does increase beyond a certain point production is worth the shot. Let me explain.

Off course a nice setup off disks in RAID10 would be a very good option. But as mentioned these are blade systems. With only 2 drive bays and 256mb integrated raid cache. To increase performance we are currently testing raid-0... as you can imagine raid-0 is way more off a risk then delayed writes would be.

Second argument is the way AppV-5 works. It creates some kind of 'repository' on the local drive with all published application, and every user that logs in to the server gets a symbolic link to every file in the repository of those apps the user is entitled to. With some of those applications exceeding 10.000 of files this means an insane amount of very small writes every login. Perfmon is showing an increase of disk queues and write latency at shift changes, making the user login process taking longer than calculated. Because this application information does not need to be persistent imo it would not be that big of a risk pushing this to the RAM with a delayed write off a couple of minutes. Most of the user profile data is redirected to the fileserver, so the user files' are not affected any different in a power-outage with or without the use of FC. In general these systems are supplied with APC'ed power with a current battery runtime of 1 hour. The biggest concern I have with FC are driver failures or some-kind of split brain situations making Windows crash/bsod. But at the other hand systems are published with SCCM so when a server does crash it is up and running again in a couple of hours. Because this are XenApp 6.5 servers the are scheduled to reboot every 3 days so I am not that afraid of the memory leaks mentioned in the forum.

The SSD concern I have is a result of the described AppV behaviour and the way an SSD performs its writes (block whipe/write). The result of this method the only case a SSD's can not really outperform 15k 2.5 SAS6 disks is when performing small writes. Off course the overall system performance will benefit from SSDs at the end, but not the login process itself which is more off a problem than system performance in general. Hence the writes and having 50 or more users logging to a single servers every day - 24x7 an SSD will wear out its write-cycles quite soon. So therefore we would have to use pricey SLC based SSD's and still end up with almost no result in decreasing the login time.
dustyny
Level 8
Level 8
Posts: 118
Joined: Sun Sep 02, 2012 12:54 am

Re: Uses cases server 2008r2

Post by dustyny »

I still wouldn't recommend Fancycache for your application mainly because it has a memory leak and because the product development is very slow. I don't want to insult the devs but there are similar products that are geared towards enterprise. Not to hard to find via google.
Post Reply