Comment Virtensys has the capability of being profoundly disruptive three times over; to network interface vendors, to switch vendors and even to server vendors. Who would have thought extending the internal PCIe bus outside servers could be like opening holes in three dikes simultaneously?
The technology is an I/O-aggregating box, the VIO 4000, that takes a PCIe link from a group of servers, typically blade servers, and provides a shared I/O resource for them. Up to 16 servers could share one 10gig Ethernet NIC (Network Interface Card), a 4Gbit/s Fibre Channel HBA (Host Bus Adapter) and one set of disk drive storage instead of each server having its own NIC, HBA and directly-attached storage (DAS).
The benefits for end-user customers are reduced component counts, meaning reduced cost and reduced electricity for power and cooling, and also less space is needed.
A general cost picture is that Virtensys would save a customer with 16 servers, needing 16 10gigE NICS, 16 4Gbit/s FC HBAs and upstream switches with 16 ports, 67 per cent on the acquisition costs and 81 per cent on the power supply cost front.
That's great for customers, but the benefits translate into disadvantages for NIC, switch and, ultimately, server vendors.
Virtensys' EMEA VP Paul Klinkby-Silver said: "We're a nuisance to switch vendors right now and we hope to be their worst nightmare." Well, don't sugar the pill Paul; tell it how it is.
First of all, a set of 16 servers would need one 10gig Ethernet NIC instead of 16. Any NIC and HBA vendor would be horrified at a 16:1 reduction in NIC or HBA unit sales to medium-sized enterprises and up. Intel, LSI and QLogic have bitten that bullet and supply NICs, HBAs etc. to Virtensys through OEM arrangements.
Emulex, Cisco, wannabee HBA/CNA vendor Brocade and others face swingeing card unit sale reductions in Virtensys accounts. What can they do? Not a lot, is the answer. In fact, it looks as if they can kiss the lost card sales goodbye forever and look to add value somewhere else.
Looking the other way, 16 servers sharing one NIC and one HBA means that the upstream switch only needs one port instead of 16. Oops. A switch vendor's business model based on per-port count costs and products just got blown out of the water. Imagine what Blade Network Technologies and other top-of-rack switch product vendors feel about that. What can they do?
They can't magic up more port count sales in Virtensys accounts. Could they supply their own PCIe network and displace Virtensys, this upstart cuckoo in their nest? Ahmet Hussein, Virtensys' CEO, says not: "It's very difficult to replicate Virtensys' technology and we hold all the patents."
In fact the Intel Ethernet card Virtensys OEMs has a built-in layer 2 switch. Oops again.
Asked about the roadmap for the product, Hussein then casually drops an improvised explosive device onto the server vendor's road: "You'll certainly see a shared memory solution from us in the future."
Pass that one by me again. A shared memory solution, as in the Virtensys product containing a big chunk of DRAM, used as the main memory by PCIe-connected server blades with no memory on them? Yes, exactly that. So, stick with me here, what's on the server blade? There would be the processor, a group of them probably, and a PCIe interface and, er, nothing else.
In that case who needs a server vendor? Stick a load of Ethernet multi-processor, multi-core server cards in an enclosure, with PCIe links off to a Virtensys box holding their memory, their shared I/O ports and their DAS. You then have a rack server without needing to buy anything from Dell, HP, IBM or whoever. All this shared server resource lowers cost, lowers energy needs and saves space. What's not to like if you're a customer?
What could the server vendors do? Build their own Virtensys equivalent? We've been there already looking at the switch vendor problem. No, the obvious alternative is, once you're certain your precious parts are irretrievably caught in the Virtensys vice, is to buy the company.
It was started as a spin-out from Xyratex in 2006 by engineers working for nothing. The first of three funding rounds, $12m, came in August 2006 with subsequent rounds of another $12m and $16m in 2007 and 2009. Ahmet Hussein became its CEO in late 2006, officially in February 2007. Paul Klinkby-Silver, EqualLogic's first European employee, left Dell eighteen months after it bought EqualLogic and joined Virtensys in October 2009.
There are ten channel partners in the UK, headed up by a value-added distributor (VAD). One of them is SGI, the Rackable SGI. Expansion into mainland Europe will follow later. The US sees a focus on Central Government with another VAD. Ditto with VADs for Japan and China, Korea and Australia and New Zealand.
A top five server vendor is expected to sign an OEM agreement in the next month or two. That would mean, we guess, one of Acer, Dell, HP, IBM and Oracle/Sun. It's probably not Dell though, as there is an Xsigo relationship there, Xsigo being an InfiniBand-based server I/O virtualisation product company.
It's still early days. The channel framework is only just being established and the OEM part of it is still nascent. Klinkby-Silver has said the channel sales people did not meet Xsigo in its activities. This looks like a classic start-up heading towards an IPO or a buyout - there's no timescale for that.
As long as the product technology delivers with no hiccups in real life, then Virtensys looks to be on a roll, a roll that's speeding up because the product customer benefits are profound. They also deliver a triple whammy to NIC, switch and server vendors, and they can't do anything about it.
Will the product perform in the real world? Can Virtensys deliver? Lots of people are going to be interested in the answer to those questions. ®