NetApp - Eisler's NFS Blog

Update on the NFSv4.1 End Game

The middle of the end game for NFSv4.1 has been reached. IETF announced today that NFSv4.1 was now  in "Last Call", which is an IETF-wide review process. Additionally, the XDR description of NFSv4.1, and the pNFS layout specifications for SCSI and OSD are also in Last Call (the layout specification for files is part of the NFSv4.1 specification).

Why do I say the middle of the end game, and not the end of the end game? The reason is that once the comment period for the Last Calls have ended, IETF's steering committee, IESG, will review the document as well as having IANA reviewing it.

Spencer Shepler and I co-presented NFSv4.1 to the SNIA Developers Conference Today

I regret not providing a pre-announcement, but I was flying in SFO this morning to present, and wanted to make sure I got to the Bay Area (the SNIA conferences is in Santa Clara, CA this week). My Blackberry skills are inferior to Gilda's and there wasn't enough time to blog from my laptop before the presentation.

The session was well attended; just under 50 people by my count. Lots of great questions. One of them was whether NFSv4.1 had any support for SSD (e.g. flash memory-based storage devices). I was a bit slow on the uptake, but then realized what the questioner had in mind. The idea is that if an NFS server (and/or the back end file system) detects an I/O pattern to data that is stored on magnetic disk drives, can NFS do anything to leverage SSD? My response is that because pNFS uses recallable layouts (effectively, striping patterns for files), a pNFS server and/or the back end file system could conceivably migrate the data from slower media to faster media (disk to SSD in this case).

This question has come up several times from customers, and obviously deserves consideration by storage vendors planning to sell pNFS servers.

I'll collect my thoughts, and blog more on the presentation as well as provide a pointer to the slide deck after my flight home to Colorado.

Brief Report from the September 2008 pNFS Bake-A-Thon

I returned home from Austin today, where a pNFS interoperability event was held at Sun's campus.

The conference room where Sun hosts NFSv4 bake-a-thons was the fullest it had ever been, with the most implementations ever in attendance. There were two three pNFS client implementations; obviously one of the two them was the Linux client. This time there were six pNFS server implementations. NetApp's Linux team brought its Linux client and server. Interestingly, the spNFS (see the slide deck and the source code) server that NetApp wrote for Linux was being used by two other commercial vendors.

No issues were found with the NFSv4.1/pNFS protocol specifications.

 

EDIT: updated count of pNFS client implementations to reflect the comment from Tom Haynes.

Heading to Austin this Week for NFS Interoperability Testing

Another NFSv4.1/pNFS  Bake-A-Thon is being held at Sun's Austin Campus. I will report later.

ONTAP 7.3 NFS Compatibility Matrix posted to NOW

EDIT: I updated this post to make it clear the omission of a client from the matrix, does not mean NetApp does not support it.

NetApp's philosophy with NFS interoperability is that we expect our NFS servers to conform with the RFCs, and thus NetApp claims interoperability with any NFS client that similarly conforms with the RFCs. We do not test every possible NFS client and every possible NFS feature with Data ONTAP. However we do test some clients, and the time has come to share this information with our customers.

At NOW.NetApp.com, our customers can find two links:

• http://now.netapp.com/NOW/knowledge/docs/olio/guides/unix_compatibility_matrix/nfs_compatibility_matrix.shtml
• http://now.netapp.com/NOW/knowledge/docs/olio/guides/unix_compatibility_matrix/kerberos_compatibility_matrix.shtml

The former is the testing matrix, with results (of all three NFS versions) between Data ONTAP 7.3 and some of the most commonly used NFS clients. Omission of your client does not mean NetApp does not support it. Again: NetApp claims interoperability with any NFS client that similarly conforms with the RFCs.

We indicate the results of testing various NFS features. NetApp has separate tests for NFS locking, exports, file operations, Kerberos, etc. and the matrix indicates whether the tests passed or failed. We also provide the list of NetApp bug numbers for client issues that we have found.

The latter link goes into more detail about Kerberized NFS testing. Customers today have a lot of choices among what Key Distribution Center (KDCs) they can use, and we have tested Kerberized NFS clients with Data ONTAP 7.3 using several different KDCs.

Our selections of what NFS clients to test, what operating systems to test, and what KDCs to test are not endorsements of any kind. They are not certifications. The selections are based on what we believe most of our customers use, and thus are intended to give customers higher confidence in Data ONTAP 7.3's interoperability.

Until NFSv4.1 ships, NetApp will be unable to use AutoSupport to definitely tell what NFS clients our customers are using. If you think we should be testing a different set of NFS clients and KDCs, let me know.