- What do you think a good "green" system administrator should do?
- What have you done?
- What are you planning to do?
- Is there any "best pratise" about ecology?
- Virtualization for server consolidation
- Paperless office as much as possible
- Shutting down workstation(s) at night
- Shutting down servers that aren't doing anything
- Pushing SunRay at the office (very cheap, very small thin client solution)
- Going to work by bike or bus
Methinks this thread should be community wiki, btw...
Pushed Virtualisation and reduced the number of servers in our server rooms by 2/3
Specified more energy efficient servers, and always opted for low power servers & server components if at all possible.
Advocated doing the same for workstation purchases, which is now being done.
encouraged people to think before they print
Oh yes, pushed out power management settings to all PCs on our campus that ensure they use as little power as possible, especially when idle (quick to sleep, etc).
reviewed power consumption of things like switches and routers and made power consumption a major part of the purchase criteria for these devices, equal with cost. (of course, power inefficient switches that generate a lot of heat but have a lower sticker price tend to eat up that price difference in terms of power consumption and cooling needs).
Interestingly enough, thinking "green" isn't terribly different from "thinking about overall TCO of devices, including power and cooling needs".
We're introducing wake-on-LAN for our Campus Grid (distributed computing system that runs on the student PC clusters on campus) so the machines only run overnight when there are computational jobs pending rather than being on 24/7
This has nothing to do with being green and everything to do with saving money as far as I can tell :)
We're running an increasing number of VMs rather than buying new kit where computational requirements allow.
Where applicable, only purchase hardware certified as EPEAT Gold. This prevents excessive packaging, brominated plastics and other nasty stuff from entering the waste stream.
Also, lease equipment that gets handled properly upon return or dispose of surplus equipment in a sustainable manner. (ie. Don't sell on eBay to some slime who is going to ship your stuff to Ghana to be smelted by children.)
Be principal in establishing working group or committee that reviews programs for coding inefficiencies
-before, or just as, the jobs go out to the clusters.
The group whould probably consist of a handful few pragmatic, not-to-judgmental devs and admins.
- If you spot, and correct, a simple oversight that makes one large batch take (1 week)*(1000 CPU:s) and get it down to (1 day)*(1000 CPU:s), then not only is that a pure time-saver, it is also a very substantial electricity saver.
- ...and, if a running time of 1 week is okay - then maybe it turns out you only need 1/7:th the number of nodes in your cluster, which would also mean quite a cut in power.
If you look at the data center from a sysadmin perspective, as others have mentioned, there are the following ways to reduce energy consumption:
- use virtualization to better utilize physical servers, getting rid of those that aren't needed
- try to make your cloud "elastic", i. e. turn on more servers during high demand, turn off servers during low demand. If possible, do the same with network gear.
- in all buying decisions, consider lifetime energy consumption, too. Also look at hardware features that help to save energy.
- check out OS features that help clock down the processor, spin down hard drives, etc.
However, there's more to the data center than servers. There will be someone who runs all the infrastructure, i. e. cooling, ventilation, lighting. Cooling easily takes up as much energy as the servers themselves. That person can use your insight to tune his/her gear. For example:
- What temperature do your machines really need? I've seen a company doing extensive research on components' heat tolerance and over time they got their data center to work reliably at 40 °C room temperature (it started to get shaky at 45, so they took a step back), saving them on the order of 80% on their energy bill. To get there, they also placed temperature sensors inside the servers and watched their readings very carefully. Of course, with tropical heat in the server room, they either had to cool it down for scheduled maintenance, or had to put on swimming trunks in an emergency (I'm not kidding, not at all).
- What usage patterns do you have? Usage is never flat, and your energy consumption should correlate with the pattern you have. If you get that to work with your servers, the facility management can follow the pattern with their gear.
- What requirements do you have for maintenance on-site? Can they turn off the lights, or turn them down? Can they switch the lights so that you can light one aisle at the time? Etc.
In complex environments like a data center, cooperation among experts is key to saving energy. Do your part.
Shameless plug: Also, do follow my site proposal on StackExchange and discuss this type of questions there, once it's up.