Despite all the hyper around Fibre Channel over Ethernet (FCoE) a few years ago, old fashioned Fibre Channel (FC) remains the dominant SAN protocol.
A report released today by technology research firm Evaluator Group shows there is good reason for that. Evaluator Group testing found FC significantly faster than FCoE with far less CPU utilization. FC also required fewer cables and power than FCoE, according to the report.
Before we get into the numbers, I want to point out that FC-centric Brocade funded the testing. Brocade sells FCoE gear too, but has been more bullish on FC while its rival Cisco has been FCoE’s chief evangelist. That doesn’t mean the results were skewed – Evaluator Group senior partner Russ Fellows said his group conducted the tests at its labs without vendor interference – but Brocade may not have released the results if FC did not come out a clear winner.
Evaluator Group used Hewlett-Packard BladeSystem c7000 chassis and 16 Gbps FC switching and HBAs on the FC side. For FCoE, Evaluator Group used Cisco UCS 5108 blade chassis and 10-Gigabit Ethernet (GbE) switching. In both cases, the storage was a 16-gig FC solid-state arrays.
The difference in response times for FC and FCoE didn’t show up until workloads surpassed 70% SAN utilization. However, FC response times were two to 10 times faster than FCoE as workloads surpassed 80% SAN utilization. FC also used 20 percent to 30 percent less CPU power than FCoE according to the report.
Speed and low latency aren’t FCoE selling points, so those results were no big surprise. A need for less cabling and power are supposed to be FCoE’s advantages, however, so it was a surprise that FC required 50% fewer cables for LAN and SAN connectivity. “This highlights and confirms the inaccuracy of the FCoE claims of fewer cables and connections,” the report states.
The tests also found the Cisco UCS required 50% more power and cooling than the HP blade with FC equipment.
The tests also determined that FC has more predictable performance with FCoE, which had twice as great a difference between average and standard deviation at 50% utilization than FC. The difference was 10 times as great with 90% workload utilization.
“If you have a high-performing application and use solid state storage, Fibre Channel is the better way to go,” Fellows said. “There is less overhead and better performance. I was surprised that Fibre Channel looked as much better than it did. The cabling and power advantage was a bit of a surprise, too.”
Fellows added that CPU utilization was almost identical when using a hardware initiator for FCoE. The test results for the report used a software initiator because that is the standard configuration for UCS, but FCoE performed better in subsequent tests using hardware initiators.
FCoE adoption for storage has been slow, for several reasons. Fellows said that while FCoE performance is good enough for many workloads, he doesn’t expect it to supplant FC any time soon. “It will continue to roll out, but I don’t think adoption will be that strong,” he said. “I think FCoE will be similar to iSCSI – it will work, people will use it and it will expand, but iSCSI hasn’t taken over the world yet.”