There's a quiet battle going on for the datacentre network -- or perhaps not so quiet.
This week, at NetEvents Asia in Phuket, Thailand, before an audience of technology press and analysts from the region, datacentre equipment vendors fought a verbal battle to upgrade a fictional datacentre. The plan was to outfit it -- virtually of course -- to make it capable of running orchestrated virtualised workloads, which generate very different traffic patterns from compared those a traditional datacentre is designed for.
Ian Keene, analyst at research firm Gartner, acted as CIO of a company planning to move its facility to be cloud capable. Vendors attending the symposium -- Arista Networks which makes high-speed top-of-rack and chassis switches; network and storage switch vendor Brocade; datacentre switch vendor Extreme Networks; and general IT vendor HP -- then pitched for the contract to upgrade his datacentre.
Keene explained the problem and the vendors made their pitches. Arista VP Douglas Gourlay said that Keene's company's IT dept was not innovating or creating value for the business. He suggested that Keene should ask any vendor of datacentre network fabrics if their technology was open, and whether you could replace parts of their solution with products from another vendor. Trashing the Ethernet fabric claims of the other three vendors, he said: "People don't want a large, flat, layer-two network. Smart network engineers don't want to build or operate them them - vendors are trying to lock you in."
Brocade VP John McHugh quipped that "our presentations are like our networks - faster than Arista." He went on to argue that "80% of network traffic will be going east-west, not north-south as the network was designed to work. Our fabric gets rid of one aggregation layer to save power, and consolidates your edge layer to create a universal touch point for most of your datacentre devices."
Shehzad Merchant, VP of Extreme Networks, said that today, everything is converging on Ethernet with no more separate network technologies for HPC, SANs and data. "Flat networks are better, with active-active paths so that recovery time is under 50ms," he said. "You need 40Gbps for the aggregation layer, and you cannot be tied into a proprietary technology - so you need an open fabric."
Later, active-active was explained as a key part of the Ethernet fabric concept. All switches are connected to each other to that as many paths are used as needed to create low-latency, high-speed connectivity to any port.
HP director Erik Papir said: "You need to look at the overall infrastructure in the datacentre not just the network. Compute power and storage are going to be critical to enable the instant-on enterprise." He went on to say that a framework to address business needs such as cloud deployment can enable end-to-end virtualisation. "It can consolidate management systems," he said. "We are open and we work with you to ensure interoperability is not a problem."
This was all well and good as far as it went. However, it was interesting that, in the various debates, presentations and informal conversations over the two-day event, few wanted to talk about storage in any detail.
As I've detailed in an earlier blog, storage startups are flowering in order to develop technologies designed to address the growing performance gap between CPU and storage. Yet no-one wanted to comment on ways of fixing it.
Perhaps it's not so surprising that, for example, Brocade, 60 percent of whose revenues derive from storage area networking according to McHugh, preferred to talk about Fibre Channel.
Yet it's clear that the datacentre's overall performance is dependent on the interlocked technologies of storage and networking. We have interesting times ahead as the two disciplines leapfrog each other.
For more about the new networking fabric technologies, take a look at my report from last year's Storage Networking World.
Disclosure: I act as editorial director of NetEvents on a freelance basis.