How Does The Flex System Stack Up Against Cisco’s UCS?
September 9, 2013 Timothy Prickett Morgan
IBM‘s Flex System modular machines were without a doubt a reaction to the success in the data center of the Unified Computing System converged platforms from Cisco Systems. But how do the two machines stack up against each other?
According to a recent report (PDF) from Clabby Analytics that is published on the IBM Flex Systems website, there are some significant advantages with the Flex System iron from Big Blue. One of them is that the chassis allows for storage arrays to be embedded into the chassis and accessed locally instead of using external storage area networks as Cisco tends to prefer with its VCE partnership with EMC. IBM offers a midplane that allows storage to be slotted in and for its Flex System Manager to see and control it.
Another big difference is that the Layer 2 switches in the Flex System chassis allows for server nodes to be directly linked to each other at a lower latency than the top-of-rack switch that provides Layer 2 and 3 services for the server nodes in a UCS setup. The UCS box also has more limited network bandwidth, which could be important for certain kinds of heavy workloads. Specifically, says Clabby, the UCS chassis has 16 10Gb/sec links between the UCS fabric switch at the heart of the machine and the fabric extenders that hook into servers and provide virtualized Ethernet links into and out of the servers. The Flex System chassis can have four Ethernet switches in the chassis for a total of 880 Gb/sec of aggregate bandwidth. This higher bandwidth is particularly important when running virtual machines on x86 iron or logical partitions on Power iron and then doing a lot of live migration between physical server nodes of those virtual machines.
If you are shopping the UCS and Flex platforms against each other, take a look. And then also call Dell and Hewlett-Packard and see what they have to say, too. Of course, Flex System is the only converged box that is going to support IBM i, so if you want to mix IBM i and other workloads on a single converged platform, then you only have one choice.
Unless one of us can talk Cisco, Dell, or HP into joining the OpenPower consortium and putting Power8 blades or modules into their systems. Now wouldn’t that be fun? HA! I will mull that over and see what they have to say.