eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
When it comes to virtualization, whether it’s in the data center, up in the cloud or down on the desktop, Intel and Advanced Micro Devices are preparing to expand the capabilities of their processors and chip sets to better handle workloads in virtual environments.
While the recent VMworld conference focused on how VMware would build and sell an operating system capable of managing all the physical and virtual components in the data center, representatives from Intel and AMD were detailing how their processors and platform would handle the emerging virtual world.
Intel and AMD, which are the world’s two leading providers of x86 processors, have been incorporating technologies into their chips for several years to make virtualization easier. However, with the increased use of the technology within the data center and the desktop, chip makers are looking to greatly enhance those capabilities.
In the coming months, customers can expect a number of new details about how both Intel and AMD will increase the ability to create I/O virtualization and make better use of memory allocations within virtual and physical environments.
Jim McGregor, research director at In-Stat, said Intel and AMD have been making big investments in virtualization technology since switching to multicore processors. Now, the challenge is the how to deal with the rest of system as virtualization continues to expand beyond its roots as a server consolidation tool.
“The next challenge, which is even harder, is virtualizing the rest of the system,” McGregor said.
“When you partition the rest of the system, you have to [take] account of dedicated memory, dedicated I/O and dedicated storage,” McGregor added. “What everyone is trying to do is take individual servers that would have been dedicated to a specific resource or a specific application and try to put a whole rack of servers into a single server. However, we still have to be able to partition accordingly to make sure there isn’t data corruption and to make sure there is security between partitions.”
What companies need to do now is figure out the best way to partition the rest of the system. For Intel and AMD, this means not only improving how their processors can allocate memory, but creating ways to virtualize the I/O of these systems. By adding these virtualization features, Intel and AMD are also creating system-on-a-chip designs that incorporate tasks that normally would have been performed on software and bringing those tasks down to the hardware level.
In the days leading up to VMworld show earlier in September, Intel made the first move to show off what it would do with virtualization with its new processors and chip sets for multiprocessor server systems.
These Intel chips, formerly code-named Dunnington and now called the Xeon 7400 series, included a virtualization technology dubbed Flex Migration, which is designed to work with VMware’s VMotion and allows for virtual machines to move between three generation of Intel processors and chip sets, including upcoming chips based on the new “Nehalem” microarchitecture.
Its in the Architecture
In addition, Intel spokespeople were talking about the virtualization capabilities found in the Nehalem architecture. These enhancements include a piece of technology called Extended Page Tables, which should help increase the performance around memory when it comes to creating virtual machines.
Intel’s Extended Page Tables or EPT resemble a technology that AMD has been promoting with its quad-core Opteron processor called nested page tables or Rapid Virtualization Indexing.
In a regular hardware environment, the processor allows for virtual memory to map to the physical memory within a system.
However, in virtual environments, the hypervisor-the piece of software that makes virtualization possible-creates another layer of virtual memory. What AMD’s nested page tables and Intel’s EPT allow for is the creation of a box around the hypervisor that allows the guest operating systems to nest within that space and lock the memory in place. This then makes switching between the guest operating systems much faster and the technology allows for better performance, while allowing virtual machines to move more quickly between different pieces of hardware.
Margaret Lewis, AMD’s director of Commercial Software, told eWEEK that what AMD is trying to do with virtualization is to take some of the more complex tasks that had been performed in the software, such as memory allocation, and move those functions to the hardware. This type of chip design, Lewis said, eliminates some of the complexities, overhead and penalties associated with server virtualization and creating virtual environment.
“Memory handling is always expensive if you do it on a software level,” Lewis said. “The more we can do with memory handling and switching between the different virtual machines and getting the memory set up so that each VM has its own space means that it makes it easier for software to run the application.”
The next step then for AMD and Intel is I/O virtualization. When AMD launches a new server platform in 2009, the company is preparing to include a technology called IOMMU (I/O Memory Management Unit), which will allow for virtualization of the system’s I/O traffic. At the same time, Intel with Nehalem will bring out a technology called VTD, which dedicates I/O resources for virtual machines. In this case, virtual I/O technology will allow the IT department to partition and assign I/O devices to virtual machines.
While at the VMworld show, Doug Fisher, vice president and general manager of Intel’s Software Division, said customers are looking for not only how Intel can speed up virtualization through its chips, but what the company can offer through its chip sets and networking such as the ability to better allocate memory and I/O devices.
“The customer does not want to pick a platform based on one feature,” Fisher said. “They look at what we are doing holistically and can we deliver the performance and reliability across the system to meet capabilities that they are looking for when they invest in a system.”