ProLiant Servers (ML,DL,SL)
cancel
Showing results for 
Search instead for 
Did you mean: 

Power Consumption data clarification from Proliant servers.

Anton Thomas
Occasional Contributor

Power Consumption data clarification from Proliant servers.


Logically speaking, the power consumption should be the same on both the power feeds( PSU) when the server is populated with two identical Power supply units.

In one of our clients DC, the data that we received was quite different.
in their infrastructure, the feed 1 is connected to the wall and the feed 2 is connected to the UPS for all the servers ( ranging from ML570 G4 to DL380 g4)
when we check the power consumption of each indivdiual server for each indivdual feeds, there is a striking difference in the readings.
for example:
If we look at the data on a ML570, there is always a difference of approx. 1kVA from feed 1 to feed 2.

PDU Value (Amps)
Date Feed1 Feed2 Total kVA
22 April 2008 8.62 7.65 3.74
29 July 2008 9.00 8.40 4.00
7 March 2009 9.30 8.10 4.00
13 June 2009 9.40 8.40 4.09


#1 Is there a logical answer for why is this happening when it is supposed to be the same?
#2 are there any tools that we could use on our HP servers where we could monitor the power consumption by the server?
1 REPLY
rick jones
Honored Contributor

Re: Power Consumption data clarification from Proliant servers.

It does seem intuitive that if a system needed to draw N units of power and had M PSUs, each PSU would draw N/M units of power but not being a EE I'm not sure if it is a guarantee without explicit action on the part of the system.

There *can* be explicit action on the part of the system that *will* result in assymetric loading of the PSUs - since it is more efficient to run a PSU at say 80% than 40% load (broad handwaving numbers), the C-class blade chassis has the ability to power down one or more PSUs to match the required load. While I doubt that is what is happening with your G4 servers, that same technology from the blades could in theory be applied to rack-mount servers.

Someone with a EE could probably explain how, absent explicit circuitry to balance them, two PSUs could end-up 11-12% different on their power draw.

I don't know that the G4 servers had on-board power monitoring - they could be old enough that the PSUs used in them didn't have the required built-in monitoring circuits. (Many of?) the G6 servers have that ability through the iLO. I think some of the G5's can do it - it will again depend on whether or not the PSUs in the systems have the circuitry and the iLO knows how to reach-out and touch it.
there is no rest for the wicked yet the virtuous have no pillows