Standard to bridge virtual gap between AMD and Intel

Pro

A group of vendors, including Dell, is preparing a new standard that will bridge the gap between Intel and AMD servers in virtualised environments.

Chipmakers currently enable the live migration of virtual machines between processor generations. AMD already offers such technology in its Opteron processors, and Intel unveiled VT FlexMigration, a similar technology that will be added to future server processors.

Migrating workloads between processor generations poses a challenge because older generations lack some of the features in newer chips. The AMD and Intel technologies solve this problem by essentially scaling down features of the more advanced chips to match those of the oldest generation in a server pool.

Such cross generation support however is merely a good start, argued Sally Stevens, director of Dell’s enterprise marketing product group.

“The real winning is going to be a capability to move across processor types, not just future and previous generations,” Stevens said in an interview.

Although Intel and AMD are best positioned to allow for migrations between their architectures, it is unlikely that the arch rivals will be able to come to an understanding there, she cautioned.
“In order to solve the problem for the customers, it’s up to the [system builders] to provide a solution.”

She declined to further specify the technology that will be introduced, how it will work or when it will be made available, other than saying that it will be launched “soon”.

Simon Crosby, chief technology officer with XenSource, a company offering an implementation of the Xen hypervisor, agreed with Stevens that Intel and AMD are unlikely to agree to an open standard to accommodate cross-platform migrations.

“They both have their own roadmaps. Their whole game is to differentiate as much as the ecosystem will tolerate.”

He furthermore argued that workloads can’t just be moved between Intel and AMD arbitrarily. AMD’s Opteron chip for instance outperforms Intel’s processors on floating point calculations, a type of instruction that is common in scientific applications. Moving such workloads to an Intel server would result in a significant performance drop.

Gorden Haff, a principal IT advisor with analyst firm Illuminata, questioned if enterprises are waiting to have workloads jump across AMD and Intel chips as they create a single, large server pool. Instead, companies will maintain several smaller server pools that operate independently.
“I see compatibility between AMD and Intel processors as a ‘nice to have’. I’m not convinced that in the near term that it is that big a deal.”

Read More: Infrastructure