The growth of virtualization has led to a tremendous rise in network traffic between servers, commonly referred to as East / West traffic. Migration of heavily loaded virtual machines (VMs) between hosts can fill even today's 10 gigabit pipes. So, East / West traffic performance is increasingly important to keeping things running smoothly in today's data center.
In July 2013, Cisco published a pair of Performance Briefs (here and here) comparing the network performance of their UCS blade architecture to that of HP and IBM. They claim that Cisco UCS, which requires traffic between blades in the same blade chassis to leave the chassis and pass through an external Fabric Interconnect, provided faster VM migration times and lower network latency.
When we read the Cisco Performance Briefs, we wondered why Dell wasn't included in the testing. Does Cisco UCS really have "the lowest latency and the fastest virtual machine migration times", as they claim, even when compared to a Dell solution?
The testing we performed revealed that a Dell solution based on the PowerEdge M1000e blade chassis and Dell Networking (formerly known as Force10) network switches provides as high as 61% lower network latency compared to the Cisco solution:
Since a large virtualization farm might have blades installed in multiple blade chassis, we also measured latency between Dell blade chassis, and found it to still be 60% lower than between blades in the same Cisco chassis, using the well established Netperf performance tool.
How does this equate to day to day network operations? To find out, we also performed a test with a heavily loaded VM migrating between blade servers both within the same Dell enclosure, and between two Dell enclosures, and compared to the same VM migrating between blades in the same Cisco chassis. As this chart shows, the VM migration was 22% faster between blades in two separate Dell M1000e chassis, and 30% faster between blades in the same chassis, compared to the time to migrate the VM between blades in the same Cisco chassis.
In the Cisco Performance Briefs, the test methodology used in the testing is described only in general terms. In fact, a note is included for folks interested in learning what testing was actually was done: "For more information about the performance tests that produced these results, please contact your Cisco sales representative."
In contrast, our whitepaper summarizing the results of Dell's testing has the exact hardware, firmware, and software used, and detailed network topologies that were set up for each solution. Cisco's solution was based on their FlexPod reference architecture, and Dell's solution was configured similarly to ensure a fair comparison. The results summarized above and in the whitepaper are the result of dozens of runs of Netperf and VMotion on each solution to ensure repeatable results.
We hope that IT decision makers will appreciate the data presented in the whitepaper and the level of detail with which it's presented, and that they find it useful in making their decisions about which blade server solution to purchase. It doesn't hurt that, based on recently obtained price quotes, the blades, blade enclosure and blade IOMs in the Dell solution cost a whopping 36% less than the same pieces of the Cisco solution, or that Dell blades are more power efficient!