Last week VMware showed some impressive hypervisor benchmark figures at the VMworld Europe 2009 conference in Cannes. The numbers suggest it makes more sense to run applications like Microsoft Exchange or Oracle databases in a virtual machine than on physical servers.
VMware says this is because such enterprise applications are not designed to run efficiently as a single instance on modern 8 core twin socket servers. While the software works on this hardware, in various benchmark tests it worked better when the hardware was configured to run several VMs configured with the same application in a cluster configuration.
During his VMworld keynote speech, VMware chief technology officer Dr. Stephen Herrod announced new VMware benchmark results from Microsoft Exchange running in virtualised and non-virtualised environments on the same server hardware. The tests were run on a 4-socket multi-core IBM System x3850 M2 server fitted with four-core Intel Xeon 7350 processors. The results show the same 16 core server hardware could host twice the number of mailboxes when Microsoft Exchange was run inside multiple VMs than when run as a single instance without virtualisation.
Similarly, earlier in February VMware claimed a world record SPECweb 2005 benchmark for the test running on a 16-core server. One of the SPECweb 2005 tests simulates the workload of an e-commerce web site. “With 69,525 connections, [our e-commerce test result] was only improved upon by 75 more connections by a very recent 24-core system,” said a company spokesman. This test used an HP ProLiant DL585 G5 server fitted with four Quad-Core AMD Opteron CPUs and 128GB of RAM.
Herrod also announced a set of TPC-C test results showing an Oracle database running in a virtual environment scored 85 percent of its performance running on the same hardware without virtualisation. Results for Microsoft SQL Server were a little better, with the virtualised SQL Server system scoring 90 percent of the non-virtualised version. Although VMware used the TPC-C workloads for these tests, the results are not compliant TPC benchmark results, Herrod added.
Even so, with benchmarks like these, it’s hard to argue any software should be run without virtualisation. After all, besides excellent performance, virtualisation also makes it much easier and cheaper to make snapshot backups and proper disaster recovery plans.