VMTN Blog

I'm over at the Intel Developer Forum this week and blogging over at Intel's IDF blog. Over the next day or two, I'll also try to touch on some recent VMware-Intel developments, including Extended VMotion and NetQueue.

Link: IDF@Intel · RAS and virtual infrastructure: keeping all your virtual eggs safe.

Rich Brunner from VMware touched on a number of topics, but the basic challenge he talked about is the “all your eggs in one basket” problem as we build the datacenter of the future. As CPUs become more capable, virtual machine density becomes higher. Imagine a future 8-way server, with each processor having 16 cores (I said future), and 8 virtual machines on each core. That’s 8 x 16 x 8 = 1024 virtual machines on this hypothetical future piece of hardware. In this future, one memory error can crash the whole physical server, bringing down 1024 virtual machines. Yes, you will have your second hot server standing by, but it seems like we should be able to do better than crashing all 1024 virtual machines for an itty bitty memory error.

VMware's Mike D is back. He talks about how you can easily VMotion from Intel to AMD, but you should understand the risks involved before doing it on production virtual machines.

Link: A Little Truth: Mike D's Virtualization Blog: Migrating from Intel to AMD.

I was at a customer this week that told me Microsoft could migrate live between AMD and Intel processors and asked why we (VMware) couldn't. I actually get this question a lot and so I'm providing some context to the answer here. ...

Actually, in the VC config file you can put in a parameter to prevent VC from doing any checks. With that parameter you can do migrations between anything you want. Just be aware that you may shoot yourself in the foot and it's not supported as such. ... With the Xen based live migration and Microsoft Quick Migration they do not perform the check and so you can actually do the migration but your app and your OS may die as a result.

But what I thought was most interesting was this little vignette as I imagine both chip makers reluctantly putting in this masking switch -- although as Mike says later in the post, it usually won't matter on the kinds of server workloads you want to be VMotioning around anyway.

With all of that said, VMware went to the chip manufacturers a while ago to see if we could get a switch where we can turn their advanced features off. They laughed at us originally. Who can blame them. They put the advanced features in there to differentiate themselves and sell processor upgrades. We're basically telling them not to be competitive. Well, with virtualization showing up everywhere the chip manufacturers have decided to put advanced virtualization capabilities into the chips. It all started with Intel VT-x and AMD SVM. Now both manufacturers are including a switch that will allow us to turn off certain features and make the processors look like they have the same features. Intel and AMD have both added this feature to their recent processor launches. This is really great since now you can buy a new box with a new feature set, add the box to the virtualization cluster, and we'll go ahead and turn off the advanced features that would normally prevent migration.

One of the trends that VMware co-founder Mendel Rosenblum talked about in his keynote at VMworld was the benefits to virtualization of new quad-core processors.

Imagine eight 1u systems, each with a single-core CPU. If one CPU gets busy and you want to balance the load, the system has to use VMotion to move a virtual machine from one processor to another. This is a high overhead operation. If those 8 CPUs have 4 cores each and are put into a single box, now the vmkernel scheduler can do the work -- a much lower-cost operation than VMotion -- and you have better utilization and even more functionality -- for instance, a security virtual machine running on a spare core. You can start dedicating a number of cores to specific workloads. Mendel showed off a very high density system (from Rackable) with 320 cores in a single rack.

Intel's Sudip Chahal has gotten his hands on some early copies of these quad core chips, and posts some early encouraging words at the new IT@Intel Blog. 

Recently Intel launched the industry’s first high-volume quad-core processor based on the new Core 2 micro-architecture, code named Clovertown. ... Frankly, the entire team is very impressed at the scaling we are seeing relative to the Woodcrest based servers (in the 1.5-1.7x range on an average – more about our methodology later) ...

While virtualizing test and development systems and other “low-risk” systems is all well and fine, the price-performance of the new Clovertown quad-core systems should enable IT to take virtualization to an entirely new level. Many applications that may not have been good candidates for virtualization in the past due to performance concerns or concerns about high costs now all of a sudden become prime candidates for virtualization and materially help move IT towards the end-goal of an automated, modular and virtualized compute utility. While previously it may not have been cost-effective to “dedicate” a host and expensive virtualization software licenses to host a modest number of more demanding applications, that is now in the past. The new very cost-effective quad-core processor based servers (pricing is comparable to dual-core servers) combined with attractive socket based licensing policies means that virtualization for the masses is now an affordable reality.