Brocade is the acknowledged leader in Fibre Channel SAN storage networks for block storage. But, much of the growth in storage today is in file storage, or what is commonly known as NAS (Network Attached Storage). NAS storage offers a robust, scale-out, high performance storage layer that simplifies storage management. However, the underlying network infrastructure for NAS storage has to be designed with the highest levels of resiliency, availability and performance. A lossless Ethernet transport can improve performance when network congestion occurs. TCP relies on the TCP sliding window for flow control. But, if congestion causes frame loss, then TCP recovery can severely restrict bandwidth as thetransmit window size shrinks and then slowly increases again. Enter Brocade’sVCS Fabric technology and VDX switch family (VDX 6710, VDX 6720, VDX 6730and the newest VDX 8770 switch). By handling congestion at layer 2 with lossless Ethernet, a VCS Fabric can avoid TCP frame loss and the time consuming recovery of full TCP transmit window size.
For the same reasons SAN fabrics are the best practice for high performance, resiliency and lossless block IO with Fibre Channel, VCS Fabrics bring these same qualities to Ethernet making a VCS Fabric ideal to transport NAS storage traffic.
Scale-out NAS Publications
The Strategic Solutions Lab recently posted several design guides and a Scale-out NAS primer showing how to cost-effectively use VCS Fabrics with NAS storage pools.
Wayne Tull, a Principal Systems Architect at Brocade, put together a very nice demonstration video showing how VCS Fabric resiliency overcomes link, path and device failures with NAS storage traffic and VMware server virtualization. What’s very cool in this video is seeing how virtual server IO to NAS storage never stops no matter what kind of failure occurs in the data path. Take a look at the last demonstration where he fails the primary IO path while a live VM migration is going on. There is no detectable pause in migration and no loss of application processing. Imagine what could have happened in a less robust network if this kind of failure caused a corruption to a database taking many hours to recover.