A while back, Chad Sakac of EMC announced on his blog that he is looking for customers to volunteer their storage arrays to run various performance tests in addition to a piece of NFS specific BETA code for DART. Having installed the BETA code (which I’m told is basically a nas executable swap in), I proceeded to compare NFS performance results with baseline results I had captured pre-patch. In most test cases, the improvements ranged from significant to over twice the performance gain. Most of the performance gains appear to surround write I/O.
Following are the results comparing NFS performance with four different workload types before BETA patch and after BETA patch on a Celerra NS-120 with 15 x 15k spindles:
Detailed supporting data. Keep in mind the NFS patch is still BETA with no firm release date as of yet from EMC:
This looks like great stuff from EMC and assuming the code reaches GA status, it would bolster the design choice of NFS in the datacenter. Chad may still be looking test results for certain use cases. If you’re interested in participating in the tests with your EMC array, please reach out to Chad using the comments section in Chad’s blog post linked above.
Ok….I like NFS and have designed a lot of environments using it.
What I’m still trying to work through (and this affects multiple companies…not just EMC) is the much faster failover times between heads/storage processors/etc. that you get with MPIO (i.e. FC or iSCSI) vs. what you get with NFS.
I’d be very curious to know if this has kept people from going with NFS and if so the specific details.
Does the new code effect Cifs performance at all? We have found that cifs write performance dies if you have checkpoints on the file system. For instance, when writing a 1 GB file to a files system via Cifs, you can get around 90MB/s performance. However, once you begin taking snapshots, performance tanks… getting as low as <8Mb/s.