This article is more than 1 year old
Cisco punts not-so-unified rack servers
California boxes ship end of June
So much for that whole argument that networking giant and server wannabe Cisco Systems is only interested in selling a closed, proprietary stack of hardware and system software known as "California" and branded the Unified Computing System. Today, with the launch of free-standing rack servers, Cisco moved to annex (in spirit) the state of Texas, where Hewlett-Packard and Dell have their x64 rack-server factories, by launching three of its own rack servers.
This week in Boston, Cisco is hosting its Partner Summit 2009, where Wednesday it rolled out a new line of rack-mounted servers based on Intel's Nehalem EP Xeon 5500 processors. Dubbed the C-Series, the new offerings complement the B-Series blade servers (also based on the Xeon 5500s) at the heart of the California box that was launched in mid-March to much fanfare, with details revealed a day later.
Since then, everybody has been wondering when Cisco would start shipping UCS. John Growdon, director of go-to-market for the company's worldwide channels, now says that the initial B-Series blades and the rest of the virtualized 10 Gigabit Ethernet networking that comprises the UCS setup would be shipping "later in the month."
The delay between launch and shipments could have to do with Cisco wanting to get some more feedback from beta testers and to build up manufacturing capacity and marketing momentum for the product. However, it seems far more likely that Cisco is trying to match is manufacturing capacity against anticipated channel sales, and that the company is still trying to build up its UCS channel.
The launch of free-standing rack servers is intended to get more channel partners up to speed on selling UCS-ready Cisco servers and related networking, but not having to sell the entire UCS stack of hardware, software, virtualization, and systems management tools. The basic idea is that a company that might want to go UCS in the future can start with rack servers.
The only problem with this theory is that the new C-Series rack servers are not expected to ship until the fourth quarter of this year. Cisco is a big, profitable company that has deep reach into the data center and that is in no hurry - just like it was in no hurry with Fibre Channel storage or switches when it entered those markets. Announcing the C-Series blades today rather than waiting until the fall keeps Cisco in all of the buying decisions for 2010 budgets, which will start after the summer break no matter how poorly the economy is doing.
There are three C-Series rack servers. The C200-M1 is a two-socket 1U box with the standard twelve DDR3 main memory slots that such a Xeon 5500 server offers, supporting up to 96GB of main memory using 8GB DIMMs. This server has two PCI-Express 2.0 slots and room for four 3.5-inch SAS or SATA drives.
The 2U C210-M1 offers the same maximum of 96GB of main memory, but has five PCI-Express slots and room for up to sixteen 2.5-inch SAS or SATA disks.
The C250-M1 is the heavy hitter. Aimed at heavy virtualization workloads, it includes the memory-expansion ASIC created by the formerly independent Nuova Systems. As I expected back in March, Cisco is deploying this ASIC selectively in its servers, and you can bet it will charge a premium for it that's just a little less painful than buying 2.7 servers with 144GB, since the memory extender chip allows a two-socket Xeon 5500 to support up to 384GB of main memory.
As Cisco correctly points out, server virtualization is more often than not running up against memory, not CPU, bottlenecks, and the memory extension that the top-end C-Series racks and B-Series blades will have is a way to get profits.
For now, anyway. If all of the other server makers are not figuring out how to offer extended memory capacity - or haven't figured out how to do it already - then they deserve to lose business to Cisco. I think you'll see similar tech from all the tier-one server makers before too long.
Cisco did not announcing pricing on the C-Series rack servers, and it has not divulged its pricing on the B-Series blades and the rest of the UCS setup.
According to Soni Jiandani, vice president of marketing for Cisco's server access virtualization group, the racks can be equipped with the same Unified Fabric (Ethernet plus Fibre Channel over Ethernet) virtualized ports as the B-Series blades within the original California blade box. At some point in the future - Jiandani did not say when - the UCS Manager software buried in the top-of-rack UCS 6100 fabric interconnect will be able to manage C-Series racks just like it can manage B-Series blades.
Jiandani says that Cisco expects most customers to not move straight to buying a complete California system - although a few probably will, if given enough discounts. Instead, they'll approach California from either the networking or the compute side. Or, like Cisco itself, they'll deploy UCS boxes as part of a normal two-to-three-year upgrade cycle or when opening up a new data center and starting with a clean slate.
The IT shops that move to UCS from the networking side will do so to simplify their Fibre Channel and Ethernet networks as they transition to 10 Gigabit Ethernet. Others will come at it from the compute side, buying C-Series rack servers for now because of the extended memory and virtual networking, and eventually evolving to a true California setup complete with Cisco, BMC, and VMware management tools and vSphere 4.0 server virtualization.
The key challenge that Cisco is up against with the California box is the perception that integration means proprietary - that you'll have to drink all of the UCS Kool-Aid to begin going down the UCS road. The C-Series boxes, if they're priced right, will go a long way towards helping Cisco be viewed as more open.
Jiandani says that of the server buyers that Cisco has polled, more than half are using homegrown system-management tools, and Cisco can sell boxes of any kind if it requires companies to switch to the UCS Manager. So Cisco knows full well that it has little choice but to be more open.
And that will very likely be true for server virtualization hypervisors at some point, as well. ®