Wednesday, November 8, 2017

Watch / Listen To Your Extreme Favorites from VMworld

All of the VMworld sessions from this year have been posted directly to YouTube this year.  In the past they were available from the VMworld site, but being on YouTube makes it a lot easier to see your favorite sessions again, see the ones that didn't fit, or experience them all for the first time if you weren't able to attend.

One of my favorite Canadians was kind enough to put together a blog post that has all of the Extreme Performance Series sessions linked on one page.  This is the series that I have been honored to be a part since it started a few years ago and focuses on performance oriented sessions.

This year my sessions were Monster Database Virtual Machine Performance and Reducing Latency in Enterprise Applications Caused by Hairpinning with VMware NSX.  The first one is about running the largest possible virtual machines with Oracle and SQL Server databases.  We have cool graphs and best practices included.  The second is about how virtual networking with our NSX product has some performance advantages over physical networking in some use cases.

(A quick note for my mom - it would be cool if you could just click on the two links for my sessions.)

Thursday, June 1, 2017

Monster Oracle Database Virtual Machines

I recently finished a whitepaper on the performance of really big Oracle database virtual machines. I highly recommend that you take a look at the paper for the full details, but a quick summary is that performance of large database virtual machines is good.  This round of testing was using vSphere 6.5 on four generations of four-socket Intel based servers.  The biggest size virtual machine that could be run without using any of the hyperthreads was tested on each host and compared:

As the power of servers has increased, the performance of large virtual machines has also increased.

Another way of looking at the performance increases is to keep the size of the virtual machine the same, and move it from older to newer hardware.  Specifically if you compare the performance of the same 40vCPU VM on Westmere-EX vs Broadwell-EX you find an increase in performance of 42%:

The older Westmere-EX based server had 10 cores per socket, so the 40 vCPU VM was using all the cores.  The newer Broadwell-EX based server has 24 cores per socket, so the 40 vCPU VM fits easily.  So in addition to the 40 vCPU VM performing 42% better, there is also room for more VMs on the Broadwell-EX server.  CPU utilization of the Broadwell-EX server was just 42% when the 40 vCPU VM was fully utilized because the host has 96 cores.