Ok. Lets call this an absolutely wild ride so far. I mean, its freaking insane. I cannot remember working so hard and so fast.
First off Tiburon, our cluster software package (designed mostly for HPC Storage, and cluster like things) has been an insanely awesome trouper. It just works. And I mean that in a jaw dropping manner. It just freaking works.
Part of it may be due to the simplicity of the thing. I dunno.
New machines. Never set them up in the database. Turned them on. PXE booted. And they were up. I mean, our DeltaV’s, our compute nodes. Just up. Fully loaded. Insanely easy to manage.
And of course, a compute node is not a storage node, so a wee little script executed in the boot process figures out if its a storage node and directs storage nodes to assemble their volumes and export them, while directing compute nodes to do their stuff. Real simple. Works perfectly on pretty much all our units so far.
Yes, it makes setting up storage (and compute) clusters as simple as putting in a new unit, turning it on, and (for the most part) walking away.
Right now we are using an Ubuntu 11.04 load (storage and compute nodes … go figure), and I will try to get the Centos 5 or 6 load up. Not sure if I will have time, so I may punt on that for the show.
We plan to show at least: GlusterFS, FhGFS, Ceph, and possibly Lustre 2.x. Not sure if we will be able to do this though given the lack of time element (have to build the booth as well, write a few presentations, 3-ish PR, and other things).
But Tiburon rocks. Not Rocks as in Rocks clusters, an older Redhat based cluster management system. But I mean rocks as in makes it completely painless and trivial to set these things up.
Tiburon is available with our storage clusters … we’ve been using earlier versions of it for years, and our compute clusters. We shipped our first Tiburon-based cluster system some 6 years ago. We anticipate supporting Centos 5.x and 6.x; RHEL 5.x and 6.x; Scientific Linux 5.x and 6.x; Ubuntu 11.04++; Debian (later model versions). Not sure about SuSE … even though one of the first Tiburon clusters was built to automate a SuSE install. Open to others. There’s no reason we couldn’t launch raw hypervisors with the tool, so we might be able to integrate it into other tools (or replace some of the mechanisms of other tools with this).
The whole concept is to make the provisioning of storage and clusters so trivial (in time as well as effort), that (even complex) storage and computing clusters (with different requirements on the nodes) and clouds are very fast to boot up, and trivial to manage/monitor. Provisioning in the cloud the way many cluster distributions do it today, really doesn’t make much sense. Either distribution of AMIs (or equivalent) or actual installation of OSes on local instances. That’s money burnt and lost.