But while the Intel Xeon chips (Nehalem is the forthcoming Xeon 5500) are the foundation for the Cisco blades' scalability and performance, their newness is also a liability. "Would I venture to put my applications on a new chip before my software provider tells me it's OK? Not bloody likely," said Tom Dugan, a chief technology officer at Recovery Networks Inc., a provider of disaster recovery services in Philadelphia and a VMware user. That concern is exacerbated by the fact the company hosts customers' environments."I wouldn't use their blades because I can't be sure that, even in the virtual environment, everything my clients run would be supported," Dugan said. The UCS B Series blade server enclosure also requires shops to buy into much more than just servers. As an integrated system, the UCS consists of the B Series 5108 blade server enclosure plus one or two 2100 Fabric Extenders that connect into a 6100 series Fabric Interconnect 10 Gigabit Ethernet (GbE) Fibre-Channel over Ethernet (FCoE) switches. That, in turn, connects into one of Cisco's Nexus 7000 core or Nexus 5000 server switch, themselves new switching platforms. And while Cisco claims that the UCS delivers 20% capital savings over equivalent systems, it also assumes a lot of up-front investment.
The required payout will be prohibitive for most organizations, said Vanover – especially if IT is organized into separate server, networking and storage groups. As at many midmarket firms, Belron's capital spending on IT infrastructure is staggered such that the server group does a major upgrade one year, networking the next, and so on. "Timing is always tricky, especially if you bring in multiple groups. Most of us just don't have the money to leap everything to top of the line all at once," he said.Even if an enterprise has the money, there's practically always entrenched expertise which is resistant to change, said Tom Becchetti, a senior infrastructure engineer at a Fortune 1000 manufacturing firm whose environment includes about 30 VMware ESX hosts, Cisco networking equipment, and Brocade Fibre Channel switches. The networking staff at Becchetti's firm is "all over [UCS]," but the storage team still doesn't believe "that IP is the best thing to carry storage. There's still a holy war about what plumbing you're using." Further, Becchetti said, he's seen too many bugs with his 10 GbE equipment but has had no problems with Brocade's 8 Gbps Fibre Channel equipment. "I'm still a bit leery of 10 Gb Ethernet," he said. VMware and Cisco's mutual appreciation society
Nevertheless, UCS includes some virtualization-specific features, which may prove a powerful lure for VMware shops.
When UCS ships, it will include the next generation of VMware Infrastructure, aka vSphere, and will include Cisco's implementation of VMware's distributed virtual switch, the Cisco Nexus 1000V. Also, new VN-Link tagging capabilities jointly developed by VMware and Cisco will give Cisco networking equipment a privileged view into network traffic not currently available to other networking vendors.
Over time, VMware will extend the distributed virtual switch to other networking providers, said Shaykar Ayar, VMware's vice president of infrastructure alliances, but for now, VMware's love stops at Cisco, and vice versa. "With all the technologies we are enabling with Cisco, VMware gets to be the differentiated and privileged virtualization vendor on UCS," Ayar said. "Likewise, on our side, Cisco is the only third-party distributed virtual switch for VMware."Looking ahead, Cisco and VMware have even more tricks up sleeves, said one virtualization architect familiar with Cisco's plans. Over time, this VMware specialist, who works for a national technology consulting firm, expects Cisco to release a consolidated network adapter (Cisco-speak for FCoE adapters) that is virtualization-specific, and that "removes the need for a virtual switch altogether." Using a combination of technologies such as Cisco's VNTag protocol, VN-Link, and the PCI SIG standard SR-IOV, "you'll be able to eliminate an entire layer … and attach a [virtual machine] directly to the underlying hardware, but still be able to do VMotion," he said.
That, in turn, will give you performance "when a VM needs full unfettered access to the underlying hardware" – and will eventually provide the foundation for a centralized Cisco-based console overseeing the entire UCS infrastructure: up to 320 physical blades running thousands and thousands of virtual machines.That kind of effort is outside the scope of most IT shops. "This technology is for the biggest companies with massive infrastructure, or for hosting companies," the virtualization architect said. "For the rest of us, it's still about rackmount servers, blade servers," and possibly, "the Nexus 1000V." For Cisco shops, "the ability to control ESX from a familiar network environment will be really tough to pass on."
Let us know what you think about the story; email: Alex Barrett, News Director.