On Sat, 19 Sep 2009 10:51:29 -0700 Jo Rhett wrote: Jo> Yes, yes it's a "state your religion" question. <action>Dons flame- Jo> retardant suit</action>
Jo> After 6 years of FreeBSD and 4 years previously of Solaris I'm back on Jo> Linux at $NEWDAYJOB. Part of my responsibility will be cookie-cutter Jo> snap deployment of blade and pizzabox test environments. Jo> The obvious google searches seem to indicate that FAI on Debian is the Jo> best support platform and has the most flexibility. (although FAI Jo> clearly supports CentOS and other distros) Jo> Tell me why you'd do this. Or tell me why not. What else have you Jo> used? What did you hate or love? I'll buy a $poison-of-choice next Jo> time I see each person who provides some detail and backing about Jo> their experience. We have a slightly odd environment (a ~64 node cluster with ~350 on demand system builds a day; think of it as a non-virtualized paas cloud) that we use a combination of systemimager and bcfg2 on. We use systemimager because it comes the closest to being a non-distro-specific deployment tool that we could find, so users are not restricted to just debian/ubuntu and redhat-like systems. (we have researchers that like gentoo for example) The other main thing about systemimager is that it is *fast*. We can do node builds in 3-4 minutes in the best case scenario; typical node builds take 4-5. Overall, systemimager is a mixed bag. Parts are a bit raw, and can be fragile. We've found that if we pull out system configurator, and stick with the smallest number of images, these issues don't cause us much actual trouble. We've replaced the configuration part with bcfg2, which allows us to use a single systemimager image per (distro version, arch) pair. Bcfg2 handles any customization that is needed, in addition to any updates that have been released since the image was created, and any user-specific customization. Honestly, this is a fairly simple workload for bcfg2, so i can't offer much useful commentary on it in this situation. It fits the bill nicely for a relatively static configuration tracking some external dynamic data (user allocations of nodes). This machine gets a pretty good workout overall. We have a user community of about 20 people, with more when students are around during the summer, and the system pretty much flies itself. The reliability is pretty nice. -nld _______________________________________________ Tech mailing list [email protected] http://lopsa.org/cgi-bin/mailman/listinfo/tech This list provided by the League of Professional System Administrators http://lopsa.org/
