At issue are two as-yet unratified standards in the IEEE for data centre switching that were being defined in concert but are now diverging. IEEE 802.1Qbg and 802.1Qbh were intended to work closely together to enable physical switches to offload much of the network-intensive processing from virtual switches on blade servers and NICs. A year ago, Cisco and HP were driving the effort in a rare show of unity.
Among the capabilities of bg and bh are Virtual Ethernet Port Aggregation (VEPA) and port extension. VEPA allows physical end stations to collaborate with an external switch to provide bridging support between external networks and multiple virtual end stations and VMs. Port extension allows users to deploy remote switches as the policy-controlling switches for the virtual environment.
Both initially used a tagging scheme, based on Cisco’s VN-Tag technology, for frame-to-VM identification and to run multiple virtual switches and multiple VEPAs simultaneously on the endpoint.
Together, the 802.1Qbg and bh specifications are designed to extend the capabilities of switches and end-station NICs in a virtual data centre, especially with the proliferation and movement of VMs. They can still do that - but bh’s role is being broadened beyond the data centre to the service provider network, and a new tagging mechanism is being defined for that purpose.
According to HP, this breaks the spirit of the original bg and bh work and forces switch and NIC vendors to implement two tagging schemes instead of one to support both bg and bh.
“Cisco decided to change the scope and it broke a lot of the cooperation,” says Paul Congdon, CTO of HP Networking. “Where we might have had a vision, we now really see two (specifications) without commonality, which is really unfortunate.”
Hogwash, says Cisco.
“Cisco did not drive this,” says Joe Pelissier, technical lead at Cisco. “This additional capability was driven by feedback from the (bh) committee during the normal ballot process. Of course, Cisco enthusiastically supports providing this capability in bh as it greatly enhances the versatility of the technology, extends its longevity, and enhances its usefulness to our customers.”
Provider networking has broadened its scope into the data centre, Pelissier notes, by helping to address VLAN scalability in large multi-tenant or cloud-based data centres. It is important then that new data centre technologies - like bh - accommodate this trend, he says.
An additional benefit, Pelissier says, is that enabling bh to operate using provider networking technologies in the data centre “by default” enables its operation in traditional provider networks.
Moreover, an additional tag was needed to support multicast traffic – the initial tag in bg specifies unicast forwarding, Pelissier says. Also, it was always the intention of the bg and bh working groups to support two tags – one for VEPA and multichannel, the other for port extension.
“Whether or not we create a new tag… just wasn’t that big a deal,” Pelissier says. “We needed a new tag anyway.”
And broadening bh for service provider data centre/cloud networking doesn’t negate the standard’s role in the data centre, Pelissier stresses.
“The main intent is still for data centre use,” he says. “But the original approach didn’t meet the intended scope.”
HP’s Congdon says the data centre focus of bh has been stripped from it.
“It’s not a data centre solution,” he says. “It was originally a data centre solution, but now it’s an anywhere-type solution.”
Congdon admits that bh could still be used for port extension in a data centre environment. But a new and different tagging scheme will make things a little more troublesome for switch and NIC vendors, and users, in that it will require additional silicon or software, operational oversight, and cost.
“We don’t really need an alternative,” he says. “The change in direction makes it more painful for people to support both. It’s much more significant. And it’s quite possible there won’t be as much port extension for the data centre.”
HP and Cisco are still together on discovery protocols for both standards, however. Pelissier says that work, which a year ago was still on the to-do list, is “99% complete.” Some of it is based on the existing Link Layer Discovery Protocol (LLDP), others on new or enhancements to existing discovery protocols.
Both bg and bh are expected to be ratified towards the end of 2011.