Cisco's Unified Server Takes Memory to the Max


A month after Cisco (s CSCO) unveiled its Unified Computing System, it has finally released pricing, processing power and memory details. The bottom line is this: the performance of the servers and overall system seem to be in line with competing products from HP (s HPQ) and IBM (s IBM) built on Intel’s latest Xeon 5500 chips, but Cisco’s offerings have more than twice the memory. That’s essential, because the goal of the Cisco box is to take virtualization to the next level.

Each system maxes out at 320 blades that have access to 384 GB of memory, compared to about 144 GB in a typical high-memory server configuration. Cisco’s blades use that extra memory to virtualize as much as possible. In addition to virtualizing the hardware so software can be abstracted, Cisco is now trying to virtualize the networking interface so the hardware can talk to whatever device it needs to on the network without manual intervention from the IT staff and a lot of cabling. In the process, Cisco has abstracted the firmware and network cards associated with each of its blades.

Adding memory helps the blades (and the overall system) keep track of a highly virtualized environment, and maintains performance in line with existing server hardware. In general, the more memory you add, the more it slows down the processor, because the processor has to figure out which memory card to get data from. To counteract that problem Cisco has also built its own chip that works with the Intel CPU to access all of that memory without taking a performance hit.

Cisco’s innovations around virtualization with the Unified Computing System are a big deal for data centers dealing with an ever increasing mess of servers, switches, cables and storage equipment, because it offers all of these units in one package. Whether or not Cisco can sell this box to traditional IT buyers is another matter, but the offering itself looks pretty intriguing.



its end to end cisco.

That means I have to buy cisco networking / servers / management. Oh I get twice as much memory? wow.

Then tell me how much the support costs..


its waste of money …………….why dont people just write distributed app on commodity hardware based on software grid platform (lots of open source option ) ………..there is lot what level you can achieve with sw based grids ……does high end hardware even matters !! Believe me no needs high end commodity for high performance computing and redundancy ………look at hodoop , mosix , beowulf ,gridgain , globas etc

Comments are closed.