Superior Performance for FC Vs. FCoE in SSD Environments

As enterprise storage users move forward with solid-state storage to accelerate business applications, questions regarding which storage networking technology provide the best results have arisen.
 
To answer these questions, Evaluator Group Inc. completed testing of storage networking connectivity between blade servers and solid-state storage, evaluating FC versus FCoE.  
 
Testing focused primarily on network performance and its impact with solid-state storage environments. The goal was to understand the impact of storage connectivity on high-performance, enterprise applications as customers adopt solid-state storage, particularly in virtual server environments.

The resulting validation found the FC environment provided 2 to 10 times faster responses as workloads surpassed 80% SAN utilization and 20-30% less CPU utilization than FCoE.
 
The testing also noted differences between the predictability of response times of FC and FCoE with increased SAN saturation. Response time predictability is important for many applications and particularly for transaction processing applications. Results showed that at a 50% load, FC provided twice the level of response time consistency, or 50% less variance than FCoE. As workloads increased beyond 50%, the variance continued to grow at a faster rate for FCoE than it did for FC, growing to more than ten-fold compared to FC.
 
“Solid-state storage has become a strategic purchase, with businesses demanding high-performance storage to deliver predictable application results. Storage networks and protocols continue to play a significant role in delivering application performance,” said Russ Fellows, senior partner, Evaluator Group. “Our testing found that heavily utilized solid-state storage environments function at their highest levels with FC connectivity.  Additionally, FC delivered superior availability, while reducing cabling, complexity and other management costs.”
 
Evaluator Group’s IOmark suite of workload tools were used to generate application storage workloads that recreate actual storage workloads. The workload environment scaled to three NL Video rendering instances at 250MB/s each, 126 virtual database applications, 126 Exchange email instances and 560 virtual desktop users. These workloads were run on blade systems from two enterprise vendors, using either FCoE or FC storage connected to 16Gb FC flash storage. Within this multi-vendor environment, it was possible to isolate the storage networking protocols, including targets and initiators, as the primary factor for any differences in performance and other results.
 
The project was funded by Brocade Communications Systems, Inc.

Full results report (registration needed)