NFS Best Practices: Networking Matters
IP storage is definitely the directional choice for many, if not most organizations. With storage now moving out of just pure storage management and into network design, this creates a unique need for systems administrators and designers to become much more intimately involved in the architecture of networking for their storage environments.
More Than Just Disks Now
As storage has evolved over the years, we have seen increases in features and performance improvements that are making innovative changes in what is happening both inside the storage platforms, plus how we access them.
IP storage, once regarded as experimental, is now widely deployed and preferred for the most part when organizations evaluate the strength of the platforms. Many storage administrators are making the jump to leverage higher line speeds (10Gb/40Gb) and more capability to aggregate network links in order to ensure there are less opportunities for bottlenecks at the network layer.
NFS has been around for some time. In fact it has evolved greatly since its inception out of the roots of version 1 in Sun environments. Now at NFS v3, the protocol has become widely used and in conjunction with VMware storage environments, we have seen some excellent features embedded.
What’s the Deal with Jumbo Frames?
This one is a challenge for some network infrastructure. While increasing the MTU (Maximum Transmission Unit) size can allow larger data packets to traverse the network, there must a common MTU across the entire path to prevent fragmentation.
Many network switch products are capped on their MTU size which will create some challenges with packing the most data for your dollar into each packet. Some storage hardware has fixed MTU sizes which will not allow for adjustment to match your network settings also which can be a limitation.
When it all comes down to it, a well architected jumbo frame deployment will be an important part of a robust NFS storage solution.
Built to Fail…No Really, it is
When we design our networking for NFS deployments, it should always, I repeat always, be architected to fail. And by that we mean a graceful failure in the event of partial path loss in the network environment.
NFS implementation in vSphere storage are fully able to leverage multi-pathing to reduce the very worst case single point of failure. There are many ways to deploy the environment for both performance and failure tolerance which is key for both the network and storage administrators to be on the same page with.
Learning from one of the Best – Cormac Hogan
If you aren’t already familiar with Cormac Hogan, he is a phenomenal resource for storage architecture and troubleshooting in VMware environments. Cormac also recently penned a book titled Essential Virtual SAN along with co-author Duncan Epping.
Cormac has posted what looks like the beginning of a series on NFS which has some great tips for networking design and we hope to see the next set of posts to help understand best practices for designing NFS infrastructure with VMware storage environments.