HP Claims Victory In Virtualisation Standards
Two standards are on their way for network virtualisation in the data centre. There’s an open alternative to the proprietary parts of Cisco’s UCS, says HP Procurve’s CTO
Virtualisation in data centres has sparked a six-month standards battle between Cisco and the rest of the industry – which is just now coming to a conclusion, according to an HP expert.
The move to virtualise servers and switches in data centres has always promised to blur the boundaries between networking and computing – and Cisco sparked off major arguments with its Unified Computing System (UCS), announced in March – in which the network giant plans to sell blade servers and converge the whole data centre.
Everyone is doing virtualisation, of course, but till this year, no one much considered the impact on the network. Hypervisors on the servers are starting to manage functions – such as load balancers and firewalls – that were previously in the network domain. Effectively, while virtualisation aims to treat the data centres’ servers as a single virtual server, it also needs to treat the network switches as a “virtual switch”.
Exposing network functions
That could waste a lot of CPU power, and it could be inefficient in terms of network resources, if the hypervisor doesn’t get a good handle on the network equipment. It could also reduce the visibility and control that data centre managers have always needed to have, over their networks.
The answer, of course, is to make network functions more visible to external management, so hypervisors and staff alike can see and manage what is going on in the virtual switch. That means data tags, which travel with the network traffic, and link the virtual switch to the actual network hardware.
“The problem’s always existed – but it was never on the agenda for the IEEE standards group till Cisco brought it up, when they brought UCS into the picture,” Paul Congdon, chief technology officer of HP’s Procurve networking group, said to eWEEK Europe on a visit to London.
Congdon is at the centre of the standards argument, as a vice chair of the IEEE 802.1 standards group, which handles internetworking and bridging. And, as he tells it, Cisco followed up its UCS announcement with a proposal to the IEEE for a standard based largely on its UCS ideas.
Centralised versus the edge
But, while Cisco favours a centralised approach – as anyone familiar with the company would expect – other vendors want to put the intelligence in edge devices on the network. “Cisco brought its proposal to the data centre group – and we brought it to the forefront and worked on it together, to come up with a solution that is more attractive to everyone.”
HP has led the group thrashing out a counter-proposal in weekly phone conferences – and this month the end result, Virtual Ethernet Port Aggregator (VEPA) has gone to the IEEE for consideration, alongside Cisco’s ideas.
“We want a natural evolution from where we are today, instead of a fork lift upgrade,” Congdon said. “VEPA changes how a virtual switch – within the hypervisor – forwards traffic, so it can be exposed.”
Server vendors including IBM have been involved, along with hypervisor vendors and other makers of NIC cards. Their goal has been to add the tags without messing up the basics of the standard Ethernet frame. “We spawned this activity through a groundswell in industry,” said Congdon – essentially, a Yahoo group did the offline work.
