When it comes to IOPS performance, VMware has its proverbial panties in a bunch. Plagued by the public perception that IO-intensive workloads don’t perform well in a virtual machine, the company is on a mission to prove otherwise, throwing tremendous amounts of engineering resources at the task.
With vSphere 4, VMware is publicly stating that a single virtual machine can drive an outlandish 300,000 IOPS, up from 100,000 IOPS with ESX 3.5. Then, at the vSphere launch last week, VMware CTO Steve Herrod told the audience that his engineers had just broken the 400,000 IOPS mark that very morning.
Clearly, VMware cares about IOPS.
Just how big of a number is 400,000 IOPS? The average Oracle database only generates about 1,200 IOPS, and the biggest-ever Oracle database was clocked in at 250,000 IOPS, said Scott Drummond, VMware Group Manager, Technical Marketing. VMware’s 400,000 number was achieved using an EMC storage array with a combination of enterprise flash memory and 600 spinning disk drives.
Drummonds conceded that ESX hasn’t always been an IOPS king. Dating back to ESX 2.5, IOPS was sub-par, but the problem was alleviated in ESX 3.0 and rectified in ESX 3.5, he said.
Nevertheless, Drummonds said he still hears the poor IOPS comment from customers all the time, no matter how modest the deployment.
“I was with a small customer in Australia recently, and he said to me, ‘Yeah, but you guys have trouble with IOPS,’” Drummonds said.
Until those comments stop coming, expect VMware engineers to continue trying to prove their case.
“We just want to put it to bed,” he said.