What’s VM stall and why does it happen?
VM stall is often encountered once users have virtualized the “low-hanging fruit” of servers and applications without high performance demands. Often, with around 25% of eligible servers virtualized, users begin to run into complexity that slows the pace of their adoption.
Thus, though server virtualization is obviously here to stay, its use is not as widespread as the technology’s high profile may make it seem. According to IDC, for example, only 19% of the x86 servers shipped in the first three quarters of 2010 were virtualized.
Opinions on why VM stall happens vary, but storage -- not surprisingly -- is a common factor, specifically visibility into disk resources. Maish Saidel-Keesing, a virtualization administrator for a technology company in Israel, estimates his environment is somewhere between 30% and 50% virtualized with vSphere. He declined to name the vendors involved, but said his company will hopefully increase its virtualization deployment rate after switching out an underperforming storage subsystem for a new one from a different vendor.
That is, of course, provided Saidel-Keesing can find good tools to monitor that performance, especially since the new storage system uses the NFS protocol rather than block storage. “I think it was only in the last release [of VMware’s CapacityIQ tool] they managed to get NFS statistics per virtual machine, or capacity analysis for NFS statistics,” he said, and third-party tools are slowly but surely following suit.
Others struggle with interoperability between application and virtualization vendors. “For my clients, the stall issue isn’t really around tools or management of the environment but vendor support,” said Matthew Liebowitz, a solutions architect at Kraft & Kennedy Inc., a Manhattan-based Microsoft and VMware channel partner, for example, the public spat between VMware and Microsoft on the combination of Exchange and VMware HA. Others point to the ongoing saga that is Oracle support for its applications on VMware.
Politics as usual
Now that Intel and AMD support virtualization at the silicon level, there are few technical issues preventing users from virtualizing highly mission-critical workloads on VMware, according to Tony Iams, vice president and senior analyst with Ideas International, Inc. “The barriers are organizational,” he said.
Vendor management is at the heart of pushing virtualization further into enterprise environments, said Chris Rima, supervisor of infrastructure systems for a utility in the Southwest whose environment is approximately 93% virtualized. But, he said, this is no easy feat.
“Identify who those people are, the technical people who’ll tell you what’s going on down to the packet level exactly what’s going on, and then [just let them] argue over the phone [until you] finally get the truth,” he said. “Make no mistake, if you go about it any other way, trying to do your own research or opening first-level tickets or trying to just get one vendor involved, you’re not going to get to an answer -- it’s going to take you months with all the finger-pointing. “
VMware and its parent company, EMC Corp., have banded together with Cisco to try to take on some of this work themselves through a joint venture called The Virtual Computing Environment (VCE) Company. VCE produces pre-integrated products called Vblocks that are meant to take a lot of the finger-pointing and guesswork out of building virtual infrastructures.
Rima said he’s going to stick with his current production environment for most applications, but is evaluating a Vblock for virtual desktop infrastructure for this reason. “We look at everything from a labor perspective,” he said. “[VCE is] a brilliant setup -- if it works.”
However, some early adopters of the Vblock report that integration problems persist even in that environment, when it comes to meshing the pre-built Vblocks with internal systems.
Beyond vendor management, Rima credits his highly virtualized environment on limiting supported OSes to Red Hat Enterprise Linux (RHEL) or Windows 2003/08 only, an emphasis on availability, and a well-designed storage and network infrastructure.
“You have to have active-passive network connections, or you have to have load-balanced active-active network connections; you need to ensure you’ve got fault tolerance in your network [for virtualization to work],” Rima said. “We test that once a year and we actually go through, identify half of production, and we pull cables. We see what stays up and what doesn’t, we remediate.”
Beth Pariseau is a senior news writer for SearchServerVirtualization.com. Write to her at firstname.lastname@example.org.