Lustre's future
By joe
- 4 minutes read - 845 wordsI’ve written here in the past about this. I have concerns as we have multiple customers using Lustre, and the official roadmap for support/releases for Lustre is anything but assuring. Moreover, it completely forecloses upon independent appliance makers using Lustre without blessing from a competitive/engaged Oracle … it is left to the reader to decide whether this will or will not happen. As I had noted before, this throws a wrench in the works of the smaller fry like Terascala. John at InsideHPC noted the same thing, and was somewhat more blunt in his analysis. His analysis, that the letter was effectively a coat of paint, seems to be something of a consensus. Terascala is ostensibly a competitor of ours, specifically w.r.t. our siCluster product which does GlusterFS, Lustre, and soon Ceph, FhGFS, and TwistedStorage. Terascala’s issues are different from DDN’s, as the latter has hired a number of Lustre developers and doesn’t need much support from Oracle going forward. Terascala relies upon an open source and publicly supported Lustre, where they can report errors, and get assistance from Oracle on high priority problems. I am not bashing Terascala. I am pointing out that Oracle has largely indicated that it has foreclosed on this practice for non-Sun gear going forward. That is, Oracle will no longer actively support Lustre on gear other than its own. Terascala’s business model fundamentally rested on Sun providing this support. This is why I thought that they will have issues going forward. I am sure they disagree, but Oracle put the writing on the wall on this … they intend to leverage Lustre to drive their business. Not Terascala’s. Or any other Lustre appliance vendor. So how do we claim our business model is not affected? First, Lustre is an option on siCluster, we have others, so if we lose the ability to offer it going forward, it isn’t a great problem (not that this would occur, I’ll explain below). Second, we are not dependent upon Oracle/Sun for support. How is this possible?
Because we lean on Clusterstor for support. They have the expertise, they can fix bugs, do ports, etc. They will support Lustre, we will support our units. Clusterstor have a vested interest in Lustre support, this is what they do. So they specifically note on their web page that they will support SLES (dropped by Oracle), RHEL, Centos, and others. I expect Oracle to drop all but OEL support over time. I expect Clusterstor to pick those up. Unfortunately, Terascala’s business model probably doesn’t work well if they engage Clusterstor as the provider of support backend. They are claiming a deep relationship with Sun/Oracle, and my view is that if they can make that work, more power to them. Given the competitive nature of the offerings, and the minimal differentiation (Terascala does what many do here and resells LSI units), I rather doubt it though. I don’t think there are many other Lustre appliance vendors out there. DDN may be one if you want to call them an appliance vendor. I see Panasas trying to exploit this uncertainty by pushing their own proprietary and expensive stack. Performance is good for them on some apps, but head to head against other very fast systems, maybe, not so much. I see DDN in cruise control, bringing big vendors with it. It has a Lustre offering, and it has Lustre developers on staff, with hard capabilities in this regard. HP and Cray signed on to this. I am surprised though that HP hasn’t leveraged its IBrix property yet. IBM’s GPFS is also used by DDN. They are hedging their bets, as no one knows precisely what Lustre’s future looks like right now. I could go on, but the point is, fundamentally, that Lustre has a somewhat uncertain future, being guided by Oracle. This uncertainty is unfortunate, and is opening doors up to competitors. Companies with sane support mechanisms won’t have trouble with current Lustre offerings, and companies that critically depend upon Oracle’s good graces are likely to be pulverized. I can’t say who, but we are well aware of an appliance company outside this space, which has a critical dependency upon a recently acquired Sun product, and they are making plans for transition now. Their user base is generally supportive of this. Lustre beyond 2.0 … it might fork. It likely would. As there are some large constituencies who want to use it, and are fearful of Oracle control. We can do Lustre, GlusterFS, and others. Even simultaneously, on the same storage units. From our perspective, the uncertainty is an annoyance for our customers and for us, but it is not an existential crisis, nor is it a case of business process uncertainty. We can ameliorate all of that. And we can support this. So if Lustre forks, or if we need to maintain a long term Lustre support, we can do this. Lustre has a future, but the near term is going to be bumpy, and the folks without a good plan for navigating the bumps … well …