Intel and AMD talk virtualisation

AS VIRTUALISATION continues to find uses for itself on top of being a server consolidation tool, AMD and Intel are both racing to cash in on the hype and bring the technology to a processor and chip-set level.

According to EWeek, Intel and AMD spent much of their time at Vmworld waffling on about how to adapt their hardware to keep up with the workloads of virtual environments.

First up on both companies' 'To-Do' lists; I/O virtualisation and making more intelligent use of memory allocations in both virtual and physical environments.

AMD will be launching its new server platform next year which it says will come with IOMMU (I/O Memory Management Unit) to virtualise the system's I/O traffic.

Intel, meanwhile, hopes to use Nehalem to wheel out technology dubbed VTD allowing techies to partition and assign I/O devices to virtual machines.

But improving how processors allocate memory is just one obstacle. Chipzilla and AMD both know they need to think how to virtualise systems’ I/O and by so doing, making SOC designs to put virtualisation tasks on the actual hardware as opposed to it being on software.

SpIntel certainly seems keen to show off Nehalem’s new virtualisation capabilities and flex its muscles by putting Dunnington’s virtualisation tech ‘Flex Migration’ on show.

Flex Migration apparently works with Vmware's Very own Vmotion which lets virtual machines switch and swap between three different generations of Intel processors and chip sets, past, present and future.

On Nehalem architecture, Chipzilla is also touting something called Extended Page Tables, which the firm reckons will up performance around memory for helping to make virtual machines.

AMD are following a similar route, with Rapid Virtualisation Indexing, or Nested Page Tables (NPT), on its quad-core Opteron processors. Normally, the processor would let virtual memory map to the physical memory on the system, but in a virtual environment, the hypervisor would set up an additional layer of virtual memory.

Both NPT and EPT box the hypervisor in, letting guest OSs run inside it and keeping the memory in place. What this practically means is that flipping between guest OSs can be done faster and that virtual machines will be able to transfer between various bits of hardware faster.

Virtually progress. µ

L’Inq
EWeek