Cisco's 3-ring circus: Xsigo CEO on bait and switches
OEM keepaway game
Blocks and Files: Cisco's John Chambers has EMC, HP and NetApp dancing to his tune and helping to sell his UCS servers and networking gear. But how does that work?
Xsigo CEO Lloyd Carney has an interesting take on the situation.
Xsigo makes director switches that enable up to 250 servers to share a bunch of InfiniBand, 10GigE, Fibre Channel, FCoE and NAS network links. Each 10GigE or 40gig InfiniBand link going to a server is carved up by the Xsigo driver into 40 virtual cables, each one of which can be assigned to a virtual machine in the server and link it to, for example, a storage LUN. The net effect is a radical reduction in server I/O cabling with 40 individual cables for NICs, HBAs and so forth, reduced to just two: one for data and the other for resiliency.
The immediate cabling infrastructure downstream of the servers in terms of switches is also reduced. SalesForce.com reckons it saves $1m in cabling and cabling infrastructure costs for every set of six Dell servers by using Xsigo's director.
What an absolutely terrific way to make a server vendor's servers more attractive compared to the competition – or to make a storage array simpler to connect to multiple servers than its competition. There is no effective competition for Xsigo's products; the PCIe virtualisers like Aprius, 3Leaf and Virtensys have either collapsed or been eaten. So it would seem to be obvious that Xsigo's products would soon be OEMed by server and storage array vendors, which would at the same time be casting acquisitive eyes over the business and its unique technology.
But neither of these things is true. Xsigo has no OEM deals and is not an acquisition target. Why is that?
The answer, according to Carney, is profitable but under-utilised cabling infrastructure on the one hand and Cisco on the other. But mostly it's Cisco.
If a server vendor sells you tens of cables and multiple network interface cards - NICS, iSCSI front-ends, HBAs or CNAs - then even if they are only 5 per cent utilised, the server vendor doesn't really care; it's all product that's bought and carries a profit margin. When every other server vendor is doing the same thing, why should any one of them cannibalise that profit stream by making less money by selling a Xsigo box? Exactly: no reason.
This is just typical server money-grubbing, with business sense being more important than a customer's best interests, nothing unusual.
But why won't NetApp OEM the Xsigo box and make its arrays easier and cheaper to hook up to servers than EMC's? Because it sells FlexPods using Cisco networking and Cisco servers and Cisco pays for all the marketing and integration, Carney says. Apparently it costs NetApp nothing at all to roll out FlexPod storage.
NetApp execs also have to consider the prospect of Cisco ending up buying NetApp, so they don't to upset the Chambers-led golden Cisco goose by adopting Xsigo directors and cannibalising Cisco Flexpod sales and pissing off the head goose. That rules out NetApp as a Xsigo OEM.
EMC? Ditto and ditto big time as the VCE kit uses Cisco servers and networking and some Cisco money, says Carney. Suppose EMC OEMed Xsigo switches? Cisco could then retreat from the VCE initiative. There is also the (distant) prospect of Cisco buying EMC. Parts of EMC really want to sell Xsigo switches, Carney says, but the VCE part and the Cisco-might-buy-us execs counsel against it. That's EMC ruled out.
What about HP? Carney says HP was going to have nothing ever again to do with Cisco after it gave HP the bird by bringing out the UCS servers. How times change, though. HP is now using a Cisco switch in a server bundle. HP also has its Virtual Connect product, which is not open and which Carney says is less effective than the Xsigo director, but it is HP's product and locks customers in. Scratch HP as a potential Xsigo OEM.
IBM? Carney says IBM prefers to sell pSeries rather than xSeries servers and Xsigo only works with X86 servers. Scratch IBM as a potential OEM.
Dell could OEM Xsigo's boxes; it is responsible for 20 per cent of Xsigo's sales at present on average. We could watch this space.
Cisco won't OEM Xsigo's boxes – obviously as it competes with its own server fabric products and switches.
Hitachi in Japan is a Xsigo partner; Hitachi Data Systems is not, possibly for "not invented here" reasons.
What's an OEM-seeking Xsigo to do? There's always Huawei. This Chinese competitor to Cisco, server and storage array vendor might love to have some cabling-infrastructure-killing boxes and so sell more effectively against Cisco, Dell, EMC, HDS, HP, IBM and NetApp. So if Western server and array manufacturers carry on ignoring Xsigo, Huawei might sneak up on them.
It's possible that Carney has been angered by Cisco and Chambers' actions. That would be understandable, but it doesn't mean his theories about a Xsigo lock-out are right. ®
Full disclosure, I work for HP and work with Blades and Virtual Connect every day.
I am curious how HP VC is not considered "Open" but Xsigo is?
Xsigo is a box that sits between the Server and the Network/SAN
VC is the same but happens to fit in the back of an HP Chassis.
Xsigo can connect to any upstream Network equipment
VC can as well.
Xsigo can connect to any upstream SAN environment
VC can, as long as you support NPIV.
So is it not open because it only works with HP's blades?
If HP was to partner with Xsigo, how would that change anything that either company does today?
HP already offers IB Adapters and Switches for all their Rack and Blade Servers, nothing preventing someone from using that with a Xsigo today is there?
I fail to see how this partnership would benefit anyone but Xsigo.
Not to mention its yet another thing I have to manage.
Something Cisco's UCS platform commonly uses as an attack point on the competition.
Re: Re: $166k to cable one Dell server?!?
Even if they're including the cost of the NICs and switch ports as "cabling", that seems like an awful lot of money.
Suppose you pay $1k per NIC port on the host, plus $1k per port on the switch, plus another $1k for each of the optics -- each link now costs you $4k, or you get 6 links for about $25k.
Heck, double that for another layer of switches above your ToR, and you're at $50k.
Even if you throw in the cost of air conditioning for all that stuff, I don't see how you hit $166k PER SERVER.
Like I said: hey, Salesforce.com, I'll wire up 6 servers for you guys for $500k. :)
It's not so much Cisco blocking them out, as the sheer marketing momentum behind Ethernet. There is no intelligent discussion of alternative networking protocols - that's certainly led by Cisco but supported by everyone else in the market.
Until Intel comes back to market with Infiniband, then Xsigo might gain some new momentum.
Also, it's a clever product. Needs clever people to buy it and there's a shortage of that right now.