Virtualization spec brings PCIe to wider systems
Keywords:PCIe? I/O chips? networking?
"These specifications, in relation to current system virtualization technologies, build on the PCIe protocol stack" in ways "targeted at improving virtualization performance," said Al Yanes, PCI-SIG chairman and ASIC designer, IBM's server group.
The standards allow a set of processors on multiple cards or chassis share common I/O chips on another card or chassis, eliminating the need for networking chips on every board or chassis. Thus, the approach both lowers costs while ensuring adequate performance for networking in complex server, switch and router systems. The specs also address issues including error handling and quality of service.
For example, using the new specifications, engineers can route Express signals from multiple computer cards or chassis over a backplane or cable to a separate, shared networking card in another chassis.
The value of specifications
The specifications "allow the creation of a new generation of virtualization solutions to meet increasing customer demands for application consolidation, lower power consumption, and lower hardware provisioning costs," said Renato Recio, co-chairman of the work group, who wrote the standards and a senior engineering manager in IBM's server communications group.
PLX Technology said it is working on a new generation of Express switches that support the IOV standards and could be used in a range of server and communications systems. "In routers, the chips could link multiple security processors, said Akber Kazmi," marketing director for Express switches, PLX.
"It will enable a totally new market," said David Raun, VP, marketing and business development, PLX. "In addition, the specifications will expand opportunities in servers where virtualization is pretty hot these days and being implemented in software in pretty much all systems," he added.
Startup NextIO is upgrading the proprietary Express switches in its systems to comply with the new specification. The company released two I/O chassis in April and expects to field new products based on the IOV standards early next year, including systems that will use third party DSP line cards aimed at use in edge networks for telecom service providers.
"The requirements for IOV are fairly complex, and because of the need for interoperability and systems management we expect we will continue to develop our own silicon," said Chris Pettey, chief technology officer and founder, NextIO.
The role of ASI
The new specifications do not replace all the functionality once conceived for a version of Express called Advanced Switching Interconnect (ASI) that was geared for the needs of embedded systems. The ASI effort was abandoned after it failed to gain traction in the face of other embedded specific interconnects such as RapidIO.
"While IOV can be used to meet some of the usage models that ASI was addressing, at best, it is a subset of what was possible with ASI so its adoption and applicability will be limited as a result," said Michael Krause, also co-chair of the IOV effort and an interconnect specialist, Hewlett-Packard Co.
"For example, ASI defined a series of protocol encapsulation services. IOV only moves Express and then only within a single virtual hierarchy," Krause cited.
"Meanwhile, Express has gotten broad adoption in computers, lowering the costs of components significantly below rival technologies in the embedded sector such as RapidIO," said Kazmi.
"We are seeing a number of adoptions of Express in parts of the industry we weren't expecting," Kazmi said, adding that an eight-lane Express switch can cost less than a quarter of the price of a roughly similar RapidIO device.
However, "most everyone will implement all the new IOV features in products, and that could slow the progress some people would like to see," he noted.
- Rick Merritt
EE Times
Related Articles | Editor's Choice |
Visit Asia Webinars to learn about the latest in technology and get practical design tips.