SharePoint with VMware vSphere 4.1
When I get around to it, I’ll be writing an article about what servers are good candidates for virtualization, and I was planning on using some performance benchmarks from vSphere 4.0. The folks at VMware’s VROOM! Blog just posted their first performance paper on vSphere 4.1 and it deals with Microsoft SharePoint.
The punch line is that vSphere 4.1 can support 171,600 heavy users (“heavy” as defined by Microsoft) of SharePoint 2007 on a single Dell R710 with 96GB RAM. Let’s take a closer look.
The team started with 3 VM’s – one for SQL, one for the app server and one for the web server. After a series of tests it was quickly revealed that web server was at 100% CPU at 72,600 heavy users. The host server was only at 35% utilization, but the web server (2 vCPUs) was saturated:
The way to add more performance is fairly obvious – add more web servers! When adding a 2nd web VM, the number of users supported increased by 87% to 135,600 users. At this point the 2 web servers were maxed out at 100% utilization but the host was only at 64%:
The host and other VMs seemed to still have more to give. When a 3rd web server was added, the 171,600 user limit was reached, this time with the SQL DB starting to be pushed to its limits:
So how was this possible? In the whitepaper, VMware credits 3 new performance improvments that were introduced in vSphere 4.1:
- Memory Compression
- NUMA-aware resource management
- Inter-VM communication improvements
You can read more details on these and other performance improvements in vSphere 4.1 here.
One caveat here. All of the VM’s were running on a single host. This allows this scenario to benefit from some of the VM-to-VM network improvements introduced in vSphere 4.1. I like to create anti-affinity rules in DRS for redundant application elements such as web servers, to better protect against hardware failure. But in this case it’s not clear that forcing the network traffic over the physical network would have significantly lowered the results — there was still CPU capacity on the host (85% used) and the whitepaper suggests that there was only a 1% improvement in network throughput over a multiple-host scenario.
This exercise also demonstrates the value of virtualization. Without virtualization, it would be necessary to purchase 3-5 physical servers to support the same SharePoint environment. When you add power/cooling, ports and overhead costs, it’s clear that there are very significant financial – as well as performance – benefits to virtualization.