HP, Cisco compromise on virtual Ethernet port aggregator (VEPA) standard

HP and Cisco have compromised on a virtual network switch standard called virtual Ethernet port aggregator (VEPA), which will manage communications in virtualised data centres without changing the standard Ethernet format. IEEE 802.1 will vote on the proposals this month.

Hewlett-Packard and Cisco have reached a compromise on competing standards for managing communications in a virtual data centre, with HP extending the standard to include Cisco's strategy on how to include legacy switches.

Both vendors have proposed a new industry standard for virtual network switches called the virtual ethernet port aggregator (VEPA). Despite eventually agreeing on the standard, both companies still differ in opinion on the other's approach.

However, both vendors aim to extend network manageability to the virtual machine (VM) level without disrupting the basics of the standard Ethernet frame. The vendors hope to have reached a solution by embedding VM information in a tag or within the Ethernet frame. The idea has gone to the Institute of Electrical and Electronics Engineers (IEEE) council for consideration.

HP has branded rival Cisco's model more intrusive, as it works on a proprietary packet format that modifies the Ethernet frame, leaving end users with hefty upgrade changes.

HP's proposal for VEPA changes how a virtual network switch inside the hypervisor forwards network traffic to the outside.

After a battle over standards, the vendors settled with an HP proposal that Cisco will build on. Cisco will build on HP's VEPA with an additional switch that meets Cisco's requirements. Cisco's approach is called a port extender, which adds a tag to automate the job of moving packets across multiple systems as needed.

Paul Congdon, vice chairman of the committee and chief technology officer of HP's ProCurve network technology unit, said that HP believes in a distributed network rather than Cisco's centralised model.

"A distributed approach is more flexible, as HP allocates tags for each switch instead of each virtual machine," he said.

According to Congdon, HP wanted to make a small change to expose network traffic, whereas Cisco wanted a new tag format and new hardware structures. As a result, HP's model works best with older network switches that are not compatible with Cisco's modified Ethernet frame.

Cisco declined to comment on the matter.

"Especially in these tough economic times, one system that manages everything would be ideal for the IT manager, and it is something that HP is working towards -- networking, management and processing all in one, for example," Congdon said.



Andy Bus, associate consultant at analyst firm Quocirca, said both HP and Cisco's models have their advantages and disadvantages. HP's approach creates tags for a virtual link between network interface cards on server and edge switches, whereas Cisco's approach is a larger switch that aggregates network traffic from several edge switches. An edge switch has more bandwidth, but a core switch aggregates more connections.

He explained: "Cisco extends management domain into the server, but this means it is running on the server, which is not always a flexible way of doing things. HP's runs on the switch, so you only have to move the switch instead of virtual machines, but this can also have its shortcomings.

"I would encourage both models, and as the technology starts to mature we will see which one works best for the IT manager."

Instead of choosing between distributed or centralised management of networking communications within the data centre, both forms could soon be supported if IEEE 802.1, the Ethernet technical standard committee, votes on it this month.


Kayleigh Bateman is the site editor for SearchVirtualDataCentre.co.uk.

Read more on Datacentre systems management