Power Consumption of VMs - XP vs Win8

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
Hi everyone,

I ran into an interesting observation recently and figured I share, my ESXi 5.5 server runs the following specs:
  1. 2x Xeon L5630
  2. 48GB ECC RAM
  3. 1 SSD + 1 500GB HDD VMDK storage
  4. 10X 3TB NAS storage
  5. Cyberpower 1000va UPS
The system runs a Linux VM that manages the NAS drives and handles other related tasks and a Windows VM that is for me to remote into to serve as a remote environment and manage the host (ESXi client).
The Windows installation used to be a WinXP VM (2 core 4GB ram) I setup 7 years ago and kept using (yes I know EOL was 2014), I've recently created a Win8.1 VM (4 core 4GB ram) to replace it and installed the same softwares and configurations. I have been monitoring power usage with the Cyberpower UPS and noticed that with the WinXP VM, power consumption was about 120-130 idling but after running the Win8.1 VM (with XP vm off), it's now up to 150-160. Both VMs have the same software load and idles most of the time. I have tweaked almost every Win8.1 power profile setting I can to save power but it seems to have little effect.

Just curious if anyone has experience with this and know why Win8.1 seems to take up so much more power from the wall even though it should be much more efficient at power management?

Thanks in advance!
 

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
Did you try assigning it identical CPU and RAM?
Stop being so logical... Just realized it after pressing New Thread. I've adjusted it to 2 cores and booting into Win8.1 seems to be down to 138ish at lowest. I've started Chrome etc and will monitor more throughout day. Seems interesting that simply having "more cores on tap" vs. actively used is actually drawing so much more power. These older Xeons aren't very good at power saving it seems.
 

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
So I reduced the cores on the Win8.1 VM to the same as XP, 2 cores with 4GB ram, still about 155-160. What is interesting is that closing Chrome gets it back to about 135-145 range, I've changed over to Firefox which was the browser used in the XP vm (sorry for that confusion) but leaving the browser open with just a few local tabs causes an increase (10W) in power usage even though it's just sitting there.
Seems like Chrome is worse but for whatever reason the browser instance in Win8.1 is just more active
 

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
I suppose all those could be factors, just seems like a big bump considering these are numbers for both VM in idle state. Considering the migration to tablets and ultrabooks you'd think Win8.1 is better about idle power consumption
 

Deslok

Well-Known Member
Jul 15, 2015
1,122
125
63
34
deslok.dyndns.org
I suppose all those could be factors, just seems like a big bump considering these are numbers for both VM in idle state. Considering the migration to tablets and ultrabooks you'd think Win8.1 is better about idle power consumption
It's good about going to sleep and great at management of cpu power states but those factors are gone in a vm i'd imagine you're seeing both operating systems at their worst in that regard.
 

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
It's good about going to sleep and great at management of cpu power states but those factors are gone in a vm i'd imagine you're seeing both operating systems at their worst in that regard.
Good point, although what is interesting is that Win8.1 VM's task manager actually has a line that states that it's a VM, so it's well aware of the environment whereas WinXP should be so old that it doesn't distinguish the 2. I'm curious to setup a new physical machine and do installs to test both
 

Diavuno

Active Member
I'm sitting in front of my old X8 system...

I will note that this is a hyper V box....

boot up peaks around 280w/300w
pause break before a full boot but after bios has loaded shows about 220 (lot of extra bits)
idle with no VM's? 210, with all my VM's it's 210

That's what my APC battery shows (the host is the ONLY device plugged in to this battery.
 

EffrafaxOfWug

Radioactive Member
Feb 12, 2015
1,394
511
113
What are your host BIOS power options set to, and what host power management options have you got set in ESX? Does your hardware support stuff like C1E or C6 states for the CPUs, or OS control for the power saving mode...?

In our testing we found that, especially with the big HPM changes that arrived in ESX 5.5, this makes a colossal difference to overall power savings and VM responsiveness (my predecessor insisted on "high performance" for everything which wasted an inordinate amount of power for no appreciable gain). Using P-and-C-states got us a saving of around 50W per host with no measurable drop in performance on our test suite. Twiddling power options in the guests makes zero difference, or at least as near to zero as makes no odds.

As to the VMs, my completely uninformed guess is that the W8 VM is simply doing more work than the XP one; my first suspicion would be the incessant drive indexing that newerer windows likes to do, extra disc thrashing might go some way to explaining it (are the windows VMs just sitting on a single 500GB drive though?). W8 and friends are more efficient than XP in that they have infinitely better power management support inside (i.e. they know about enhanced sleep states and whatnot whereas XP doesn't) but on the flip-side they definitely do more thrashing around on frippery than XP did. Have you tried turning off the drive indexer and all that sort o' gumpf? Going into esxtop's power view ("p" key) should be informative in this regard, even if you don't have an ACPI-aware power supply it should still give you lots of info about the processor power states.

Simply having stuff sitting in RAM shouldn't really increase power consumption - the RAM is going to be refreshed whether there's stuff in it or not. It's shunting lots of stuff into and out of RAM that uses power.

Could windows server be doing power management that esxi isn't capable of?
Out of the box, if I remember correctly I think ESX is quite conservative about what power settings it enables whereas windows/hyper-V is more aggressive by default.
 
Last edited:

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
What are your host BIOS power options set to, and what host power management options have you got set in ESX? Does your hardware support stuff like C1E or C6 states for the CPUs, or OS control for the power saving mode...?

As to the VMs, my completely uninformed guess is that the W8 VM is simply doing more work than the XP one; my first suspicion would be the incessant drive indexing that newerer windows likes to do, extra disc thrashing might go some way to explaining it (are the windows VMs just sitting on a single 500GB drive though?). W8 and friends are more efficient than XP in that they have infinitely better power management support inside (i.e. they know about enhanced sleep states and whatnot whereas XP doesn't) but on the flip-side they definitely do more thrashing around on frippery than XP did. Have you tried turning off the drive indexer and all that sort o' gumpf? Going into esxtop's power view ("p" key) should be informative in this regard, even if you don't have an ACPI-aware power supply it should still give you lots of info about the processor power states.

Simply having stuff sitting in RAM shouldn't really increase power consumption - the RAM is going to be refreshed whether there's stuff in it or not. It's shunting lots of stuff into and out of RAM that uses power.



Out of the box, if I remember correctly I think ESX is quite conservative about what power settings it enables whereas windows/hyper-V is more aggressive by default.
All valid points - my ESXi host is set to max power savings and all power saving features in my BIOS that I can find is turned on. My original config was ESXi5.5 with the XP VM built way back from VMWare server on Linux and migrated to ESXi5.0 and now to ESXi5.5.
The Win8.1 VM was built purely so I can run VMclient since I don't have a Vsphere license for web management, and power profile is set to "power saving" (I did disable sleep timer as it would prevent RDP sessions and defeat the point). I have turned off drive indexing and a couple other "idle maintenance" services/tasks and lowest I can get is 150W at the UPS. With XP vm it's about 125.
Regarding where the VMDKs are, ironically the XP vm was sitting alone on a 500GB 3.5" drive and Win8.1 vm is on a Intel X25 SSD (which you would think saves even more power).

I really am stumped other than like you said, W8 loves some thrashing
 

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
FWIW there is a fling out there for web management: ESXi Embedded Host Client – VMware Labs

Depending on your needs this may negate the need to run the windows VM altogether.
You just made my day! It does require ESXi5.5u3 (seems like there's an b and c version) or ESXi6.0 so I'll need to check those out. Glad to seem VMWare making some toys for us homelab guys
 
Last edited:

RyC

Active Member
Oct 17, 2013
359
88
28
You also need to have a non-free ESXi license for it to work currently
 

chinesestunna

Active Member
Jan 23, 2015
621
194
43
56
Haha I'm sorry!
Actually I just re-read the page, I don't see any requirement saying one needs a paid vSphere license. Only a known issue refers to problems with ESXi hosts with a paid license in fact.
The VIB itself is released under a technical preview license.