I’ve said it before that I’ve always been a huge fan of SystemImager, for super simple imaging. It has some shortcomings for config management, but those are solved with things like Chef or Puppet.
With all of the great things being talked about surrounding ‘Automated Infrastructure’, I’ll point to something insanely cool: 1,190 nodes installed from bare metal to all done in 15 minutes.
That’s One Thousand One Hundred and Ninety nodes. Completely installed in: Fifteen. Fucking. Minutes.
Pingback: BotchagalupeMarks for July 17th - 14:33 | IT Management and Cloud Blog
Now imagine one thousand one hundred and ninety nodes with Chef baked in and enough supplemental JSON to take them to serving apps, in (I imagine) the same amount of time.
# TODO: attempt this
Rocks (www.rocksclusters.org) put bittorrent into their install two years ago. I regularly reinstalled our 1000+ node cluster in this amount of time.
Systemimager was in the running for our install process for a short time. But their golden image configuration setup is a pain, and at the time, they didn’t have an install process that scaled. Additionally, using a separate tool to do configuration management is also a pain. Rocks makes it simple to set up a cluster using regular kickstart and getting the configuration you want right from the get go.
And you can install 1000+ nodes in 15 minutes too.
Big hairy deal.
Thanks,
Joe
Rocks or SystemImager, I’d say that *not* using an automated install tool for 1000+ nodes is a big hairy deal. 🙂 For those folks in the HPC world, clusters this size is commonplace. Not having worked in that environment for some years now, I don’t know how often a full install happens in normal operation.
In the web operations world, most installs I’m aware of run anywhere from 10-100 machines at a time, being added to various smaller clusters. While bittorrent-laced installs are insanely cool, you and AJ are right: the really cool stuff happens when you’ve got some real configuration management next to it.
IMHO, setting up SystemImager wasn’t a pain even years ago, but that was because there really wasn’t any alternative except for Rocks, which we weren’t going to use because it didn’t fit the working paradigm. (it seemed, if I remember correctly, if you’re using rolls for everything, you’re in good shape. if you don’t, then it becomes more awkward to manage)
Regardless: the way automated infrastructure happens in web shops these days mean having a base OS image install system (SystemImager, FAI, Cobbler, Kickstart, Jumpstart, AWS AMI images, etc.) and then having a Config Management system take over from there (Chef, Puppet, BCFG2, etc.)
Pingback: Cloud Cafe #86 - Agile Infrastructure | IT Management and Cloud Blog
Pingback: Agile Operations | Web Admin Blog
Pingback: Agile Operations « the agile admin
FAI does not useOS images, but does the installation coltrolled by scripts.
The FAI webpage is http://fai-project.org