Virtualization: Blessing or Curse?
I saw this float across my Google Reader yesterday, thanks to Tom Limoncelli. If you are a sysadmin in an environment fixing to do more virtualization, it is well worth a skim:
Virtualization: Blessing or Curse?
NOTE: this isn’t an anti-virtualization rant, more of a “things to watch out for” briefing.
Some of my take-aways:
- Sure we’ll have fewer physical servers, but the number of deployed systems will grow more quickly. As that grows so too will our systems management burden.
- As the system count grows faster, we may hit capacity chokepoints on internal infrastructure like monitoring, trending, log analysis, DHCP or DNS faster than previously assumed.
- Troubleshooting becomes more complex: is your slow disk access an application, OS, or hardware issue becomes also a potential host OS, networking/SAN or filer issue as well.
- Regarding troubleshooting: we may add another team to the mix (to manage virtualization) while trouble-shooting has an increased probability of requiring cooperation across multiple teams to pin down. Increased importance on our ability to cooperate across teams.
- Change management impacts: small changes against a larger number of systems sharing architecture snowball even more. One can add something to the base image that increases disk use by 1% for any one system, but multiply that across all your systems and you have a big new load on your filer. (1,000 butterflies flapping their wings.)
- Reduced fault isolation: as we have greater ability to inadvertently magnify increased load and swamp network and storage infrastructure, we have a greater ability to impact the performance of unrelated systems which share that infrastructure.
- The article also cautions against relying on vendor-provided GUIs because they don’t scale as well as a good management and automation framework.
Ah, the other thing noteworthy there is the ACM Queue magazine is now including articles on systems administration. (I subscribed to the system administration feed.)
