Posts tagged with x86

I applaud AMD for instituting Virtualization Ecosystem Month because it reminds us of the Three Musketeers philosophy: “one for all, all for one.” Only through our combined efforts can we meet customer needs.

It’s hard to believe that only a few short years ago, hardware assisted virtualization was not generally available on the x86 platform. The intervening years have brought an enormous amount of innovation that translates directly into economic value to customers. And we’re not done yet.

While some may be looking for the next big problem to apply virtualization, many IT organizations want to keep on doing what they’re doing, only better. What is holding them back? From a workload perspective it is usually support, performance and risk. Support and risk can be lumped together as problems that require more than technology to solve. Workload performance has technology written all over it.

It used to be a truism that you could only virtualize the smallest, least mission critical workloads. But we read every day about organizations that have success virtualizing ever larger workloads. However, there are still some workloads that don’t fit comfortably in virtual hardware. If we want to virtualize everything, we need to reduce virtualization overhead.

Overheads are an inevitable part of virtualization; each layer of abstraction introduces another step to the dance. The trick is to identify which of the extra steps takes the most time, or is performed the most frequently. And then to find ways to remediate these overheads. One step along this path is optimizing SMP (i.e., more than 1 CPU) workloads. Once you move to a second CPU, there are all sorts of interesting issues introduced. As we tested larger workloads, we noticed that virtual CPUs were sometimes scheduled when they shouldn’t be. This caused extra processing without any benefit.

After analyzing the problem, we made a change that resulted in substantial performance improvements. But we worried about the impact this change could have in other scenarios. Exiting the virtual machine to the hypervisor is an expensive operation, and could introduce latencies if it was performed at the wrong time.

This is where the ecosystem shows its power. AMD technologies like AMD-V™ and RVI help reduce hypervisor and virtual machine overheads, allowing organizations to run bigger and more mission critical workloads. I think it’s a great demonstration of how hardware and software vendors can work together to solve a problem.

Much as the Musketeers learned, there may be obstacles along the journey, but perseverance and collaboration is what brings success in the end. The entire ecosystem benefits. And the results last a lot longer than a month.

Chris Barclay is responsible for product management and technology partnerships at Virtual Iron Software. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Any claims made herein have not been independently verified by AMD. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.