Searching through the posts, I don’t see any post about virtualization in the last couple of years. With the trend toward virtualization, it seems like there would be a lot of posts about it.
Anyway we are implementing about a 15,000 client BES deployment in an VMware ESX environment with the following specifications.
OS Windows Enterprise 2008 R2
Database SQL 2008 R2
Memory 8 GB
Processor 4 cores – hosted on Dual 6 core Xeon 2667 MHz processors, 144GB of RAM in BL490 HP blades
Disk 100 GB Backend Storage on IBM XIV Storage for the OS and Bigfix application
Disk 100 GB Backend Storage on IBM XIV Storage for Bigfix database on SQL 2008 R@
4 GB fiber connection to the backend storage
My question is: is the 4 GB fiber connection to back end storage going to be a problem?
What did you search for? VMware, virtualization, and virtual should all turn up a lot of results (recent and old)…
The most important questions are not if you are using VMWare… The more important points are:
Your CPU/memory looks fine…
Are you sharing the hardware/disks for the database? If so, then you will likely hit performance issues because the BigFix database will need all the IO it can get.
In our environment, I don’t have much control of the sharing. One of the main reasons for virtualizing is to share resources so it is hard to ask for dedicated resources. I guess we will have to wait and see how the performance is and make adjustments.
The IBM XIV is supposed to be a high performance disk system.
We also have an IBM SAN and VMware. We moved even our dev environment off of VM due to I/O issues. Ben is right, getting sufficient IOPs is key to a successful deployment.
In my opinion, it is worth fighting for dedicated, local disk. Buy as many spindles of the best local disk you can afford.
If you need a lot of IOPS for a large deployment of Bigfix, (I’m running on a good high-throughput enterprise SAN and it’s fast and Bigfix performs well) don’t scrap virtualization because of this. Very many options out there. FusionIO is one of them if you really want IOPS. Faster than any local storage you’ve ever seen. You can even put a couple of their cards in your VMware hosts and use Vmware’s VSA (virtual Storage Appliance) and make the local storage look like shared (with Vmotion) storage and get over a million IOPS. Bottom line is high IOPS in virtualized environments is easy to get if you build your clusters right. Above is just one example.
It is less about throughput and more about latency due to our constant small database updates from the agent reports… We have seen really terrible performance from some high-throughput SANS because they have high-latency… Here is some more info: