So, a lot has changed in the last year. The big news is, Cisco is now in servers! I was going to wait a while, see how the dust settled, and wait to play around with them until Cisco’s voice apps were approved to run on their own servers.
Well, during that waiting period, something exciting happened – Cisco dropped the price on their servers bigtime. They are now competitive with every other major server vendor out there. This changes things!
Sure, they meet all the requirements of an enterprise class server. Good OS support, out of band management, reliability, power, etc. They also have a few unique features.
The B series servers are blade servers that go in the Cisco Unified Computing System. This system has three major advantages over anyone else out there right now. The first is that the management of the chassis is done at the top of the rack by the chassis manager application running in the 10Gbps FCoE capable switches at the top of the rack (called Fabric Interconnects) – that means no onbox management engine, and that the chassis is truly just a provider of rackspace, power, and uplink ports. This lowers the cost of the system, and makes it much easier to manage, because all 300 or so blades can be managed from one system.
The second major advantage of the B series blade servers is the uplink cabling. The chassis uplinks to the top of the rack with 10Gbps twinax copper cabling. No fiber SPF’s, no fiber runs, no multiple gigE, no Fibre Channel adapters, nothing else. Just the multiple redundant copper cables, and not very many of those. Way less cabling and adapter costs, much better airflow for cooling.
The third advantage of the B series blade servers is Cisco’s unique memory controller. It allows for 48 DIMM’s to be put into a single server. Does anyone need 384GB of memory? Probably not. But I could see populating the box with a bunch of cheaper 4GB DIMM’s, giving 192GB of memory, in certain applications. The first one that comes to mind is hosting a bunch of virtual desktop hosts in a VDI implementation.
See, the whole value proposition of Cisco’s servers is to provide the highest virtualization density possible. The limit to the number of virtual servers or desktops on a server is bounded by processor capability, memory size, network bandwidth, and storage speed. Intel ticks and tocks along with multicore processors upgrades, so that’s taken care of. Cisco is pushing the other three limitations with the Unified Computing System design! 10Gbps for network, FCoE for storage, and Cisco’s unique silicon for memory.
OK, so what about the rackmount servers? Well, they have to make some, since not every application calls for a fancy Unified Computing System. These are called C class servers (I have no idea why), and come in three flavors – 1RU, 2RU, and 2RU with extended memory.
These are pretty much commodity x86 servers, although supported by Cisco. There is a roadmap to the servers being part of the UCS, one of the boxes can have lots of memory, and that’s about it. Although Cisco ships a lot of rebranded servers currently, so by using their own boxes they get to keep their margins high for a few more years.
So, if you are thinking of getting some more servers, give your local favorite Cisco partner a shot at the business.
____________________________________
Author: Rolf Versluis
Adcap Network Systems – Atlanta and Miami
Great Local Engineers Creating Systems that Work!
Posted at Adcap Tech Tips
Related posts:
Last Updated: June 10th, 2010 |
