2

If I have some servers set to static high performance mode, is there a way to generate a report on what power savings mode would have done, if it were enabled? I am looking to generate reports of the form, on this day we used this many kWh that we didn't really need to, summed across the estate. Or an alternate method to do this?

Gaius
  • 1,471
  • 1
  • 12
  • 19
  • 2
    This is interesting. Let me think about it. – ewwhite Dec 21 '14 at 20:03
  • Can you provide more information on your servers? Model(s), operating systems, workload, etc. – ewwhite Dec 21 '14 at 21:13
  • We don't generally ask for idle curiosity, but because the information has a bearing on the question (and answer). It's to your benefit to provide as much information as possible. – Michael Hampton Dec 21 '14 at 22:14
  • Can you also explain how you're being billed for power? Is it per-server, per-circuit, are there thresholds you need to match, is this a co-location facility/data center, or are you trying to meter your utility power utilization?? – ewwhite Dec 21 '14 at 22:19
  • Bear in mind that for most virtualisation products "Static High Performance" or the vendor equivalent (I.e., disabling C States etc) is recommended best practice so your "didn't need to" may well be a questionable statement in these, and other, instances. – Dan Dec 21 '14 at 22:19
  • @dan I know that this recommendation has been made in the past because Linux's power management is so flaky, I wonder if it's any better now – Gaius Dec 22 '14 at 08:04

2 Answers2

1

I'm unaware of any current server that holds this data and provides estimates based on it's history, nor any upcoming servers that do so. The only way to get close to a real answer would be to test it yourself with a power meter over a period of time with both settings. That said I buy HP servers and their 'balanced' power profile is expected to save at least 15% of power draw over their 'maximum performance' option.

Chopper3
  • 101,299
  • 9
  • 108
  • 239
  • The ILO keeps a bit of history. This is easy to test. Run a server in both modes for a day and compare the graphs. I would have posted examples had the OP provided server model and generation information... *pulling teeth* – ewwhite Dec 21 '14 at 22:47
  • @ewwhite thanks, we have everything from G2s to G9s. I'll have a look at the ILO graphs and see if they are useful, but what I really want is the "what if" scenario – Gaius Dec 22 '14 at 08:02
0

Measure it on typical servers and calculate that on this base. And/or ask your hardware manufactorer.

Recent servers should be equipped with online power consumption measurement means and even statistics. (e.g. Dell PowerEdge x9xx or Rxxx models)

Nils
  • 7,695
  • 3
  • 34
  • 73