Why is converged I/O in data centres taking so long?
Six factors holding back infrastructure mergers
By Jim Duffy | Network World US | Published: 12:30, 26 September 2011
So if converging the I/O infrastructure in data centres is all the rage, what's taking IT shops so long to do it?
- New technology attempting to replace proven and reliable implementations.
- New equipment requirements.
- New standards and proprietary techniques to consider.
- Organisational and operational changes.
- Infrastructure management and stability.
- And, questionable benefits beyond the server and access switch layer.
Converged I/O, running LAN and storage data through the same wires and switches to reduce elements and cost, is a years-long journey though, not an endeavour to be rushed or taken lightly. IT shops have to weigh their situation carefully and know where they want to go, and how and when to get there, before embarking.
Generally, converged I/O constitutes three key elements: 10Gbps Ethernet, Fibre Channel-over-Ethernet (FCoE) and Ethernet equipped with the lossless Data Center Bridging (DCB) standard from the IEEE. FCoE, which tunnels Fibre Channel storage traffic through Ethernet, requires DCB in order to have Ethernet behave as if it had the resiliency of Fibre Channel.
According to Dell'Oro Group, FCoE realised $94 million in revenue in the second quarter of 2011, on a shipment of 210,000 ports. The research firm expects 930,000 ports to ship this year, accounting for $422 million in revenue, or about 7% of 10G Ethernet revenue.
But there are many standards and emerging standards to consider, as well as proprietary vendor schemes, when evaluating a converged I/O infrastructure throughout your data centre. Currently, those standards are in place for FCoE and DCB at the blade server and access switch level and FCoE is largely a free technology feature of 10Gbps Ethernet switches and converged network adapters.
But standards and methods for extending converged I/O from the server rack and access switch, where it is now taking hold through the core of the data center network, are still percolating. Indeed, some of these standards are competing to become the de facto technique for enabling multi-path networking and multi-hop FCoE capabilities.
Massive growth potential
Xsigo is a maker of virtual and converged I/O infrastructure products, namely its I/O Director and Server Fabric platforms. In 2010, privately held Xsigo more than tripled its revenue from the year before, so the company sees a lot of demand for and sales of converged data centre I/O gear.
"When servers ran one application per server and that application did not change, you're typically only dealing with a few network connections per server, and pretty low utilisation of those connections," says Jon Toor, vice president of marketing for Xsigo. "When you virtualise a server you're dealing with a lot more connections, a lot more workload and it creates a need for a different way of hooking things up."
Xsigo, though, pitches its products as alternatives to having to deploy FCoE to achieve converged data centre I/O. Cisco, the market leader in FCoE switches, has been attempting to undermine Xsigo's strategy and company stability.
HP, which enjoys a 20% share of the FCoE blade switch market, says more than half of workloads will be virtualised by 2012, which puts additional strain on the access network.
"Customers are deploying six to eight Gigabit Ethernet connections because virtualisation requires more bandwidth out of the servers," says Kash Shaikh, director of marketing for HP Networking. "That amount of cabling blocks airflow. (With converged I/O) you can take that down to two 10G connections coming out of the server and into the first hop switch."
In addition to server virtualisation, other converged I/O drivers are cloud deployments, infrastructure flexibility, an increasing amount of server-to-server traffic and consolidation of I/O density with virtual machines, says Shaun Walsh, vice president of marketing with Emulex. But the march to converged I/O will be gated on the amortisation cycles of IT shops and the question of who and what will manage the new infrastructure.
"They have existing infrastructures that they need to amortize over time to get the full value from them," Walsh says, which is usually three to five years.
Management of the infrastructure from both an operational and organisational perspective will have to be considered carefully as well, he says. LAN and SAN teams manage segregated data and storage networks now with likely different operating methods.
"The biggest challenge for organisations is not physical deployment, it's the policy and management deployment," Walsh says. "Sit down with the teams, make sure that they have a meeting of the minds on what the purpose is, why we're doing it and who's going to manage what segments of it."
There can be some hesitation when it comes to combining traffic from currently isolated networks.
"What are the security implications of adding Fibre Channel to an IP-driven environment?" Walsh asks. "That's always one of the big concerns storage administrators have expressed to us."