Vint, Vish, Sage and Alex at LISA 2012

Posted by Alex McDonald


LISA ’12, a USENIX conference, is aimed at system administration practitioners and researchers. The conference is a mixture of tutorials (that’s the cheap tickets), workshops (pricier) and various hallway poster sessions and Birds-of-a-Feather meetings (which are free).


Armed with a cheap(-ish) ticket, I flew in to San Diego to attend a few of the more interesting sessions and present an NFSv4.1/pNFS BoF, of which more later.


I missed the keynote on Wednesday; The Internet of Things and Sensors and Actuators! by Vint Cerf of Google due to a short early morning meeting that turned into a long mid-morning meeting. I did manage to sneak in at the back end to a packed auditorium. It was described by one delegate I got chatting to at the end of the session as “Google’s game plan for world domination; the tech version”.


OpenStack: Leading the Open Source Cloud Revolution, Vish Ishaya, Nebula, Inc.


Vish, as one of the early developers of OpenStack while working for NASA, outline the history and some of the features of OpenStack, along with how the OpenStack Foundation is organized. An OK history lesson, the technical element was pretty light, and the demo was whizzed through at great speed. The futures part of the presentation, as can be expected from a project that seems to design everything by committee through OpenStack Summits, was very light on detail.  NetApp -- and NetApp alone -- got a mention; for our contributions to Cinder (the block storage part of the solution).

Personal opinion; I’m not sure that OpenStack is ready for the big time yet. It’s a little rough and unfinished, and the storage support layer (Swift, very similar to Amazon’s S3) seems primitive in the extreme.


Ceph: Managing a Distributed Storage System at Scale, Sage Weil, Inktank


Sage Weil presented well on the topic of Ceph, an open-source object based file system that can be implemented out of commodity parts. Inktank, which Weil runs, is the company that develops & supports Ceph, and they had a booth at the exhibition.

There was a large (if slightly sycophantic) audience for his presentation, and a great number of questions afterwards from some who had or were about to implement it. For me, it had the feel & finish of a science project; lots of cool ideas, but not much by way of enterprise readiness or commercial support. It does have its attractions though, and it wouldn’t surprise me to see Ceph more widely used in the next couple of years. One to watch.


Every so often I run a Birds of a Feather at suitable conferences entitled NFSv4.1 and pNFS Ready for Prime Time Deployment BoF. It’s been an informal BoF run for quite a few years by Sorin Faibish of EMC, who has moved on to other work, so I have taken it over. The goal of the BoF is to NFSv4.1 and pNFS in an industry neutral way; each of the vendors with an NFSv4.1/pNFS server implementation, and the providers of clients – all Linux based distributions – gets a few slides to describe their product and solicit feedback from the attendees.


This year’s LISA BoF had 35 people attend it; that’s a sizable number. Doug O’Flahery of Tonian (a small Israeli development company) and I hosted the meeting and presented each fo the vendor’s solutions; EMC, NetApp, Panasas, IBM, Microsoft (yes – they have an NFS4.1 server with Windows 2012) and a number of others.


Perhaps the most important part of this is the NFS client; and many might not be aware that NetApp fund several full time Linux NFS client developers led by Trond Myklebust that have done great work over the last few years in getting an enterprise quality implementation of NFSv4.1 with pNFS ready. The major distributions of Linux are now including that work; RedHat have included support in RHEL (RedHat Enterprise Linux), SUSE are preparing to do so in SLES (SUSE Linux Enterprise Server), and we look forward to Canonical’s Ubuntu supporting it shortly.


Feedback at the meeting was generally positive if cautious. Perhaps because of the audience that attends LISA, there was a very large and noticeable lack of understanding of the features of NFSv4.1/pNFS, and how they benefit specific application and workload use cases. More general education is required; it’s not understood as well as other technologies deployed & managed by sys admins. Security is an issue; some were concerned at the level of effort required to kerberize their environments for NFSv4.1.

However, there was much positive support for this new & improved NFS, and I look forward to 2013 and 2014 seeing increasing awareness and adoption.