Measuring power
This is a long way off topic but I'm hoping that someone here might be able to share a clue... I need to measure the power used on dozens and dozens of servers. I need a breakdown by server, not a total for the room. (more specifically, I want to calculate the power cost per server per year). Added difficulty: I want to do this non-intrusively. I can't go round, yanking out cables and inserting in-line meters. So, I got a clamp meter. But contrary to my expectation, that only works if I can separate the individual wires in the power cable - not an option. So, I'm stuck. Is there any non-intrusive way to measure the power consumption of my servers? Andrew
Re: Measuring power
On 29 Apr 2009, at 09:42, Andrew Beattie wrote: So, I got a clamp meter. But contrary to my expectation, that only works if I can separate the individual wires in the power cable - not an option. So, I'm stuck. Is there any non-intrusive way to measure the power consumption of my servers? If you can't separate the conductors, no not really. There are some very expensive power distribution units that will provide per port current usage, but that's still involves pulling plugs for existing serveres. If you can at least power down the units briefly, you could insert a short stretch of cable with separated conductors. Your other option would be to lay hands on a similar server and test it inline if you need general rather than specific measurements. Most normal servers pull between .5 and .7 amps in my experience if that helps. :) - Mark
Re: Measuring power
A frend of mine has a Power Owl http://www.greenstamp.co.uk/product_info.php/cPath/33/products_id/68?gclid=COmqvdzclZoCFR4hnAodx3XzNw You clip the sensor round the mains power were it comes in to the electricity meter. I guess you could clip it round the server power cable I'd be inclined to calibrate it against a inline power meter before use... -- Michael ~~~ Michael John Lush PhD Tel:44-1223 492626 Bioinformatician HUGO Gene Nomenclature Committee Email: h...@genenames.org European Bioinformatics Institute Hinxton, Cambridge URL: http://www.genenames.org ~~~ On Wed, 29 Apr 2009, Andrew Beattie wrote: This is a long way off topic but I'm hoping that someone here might be able to share a clue... I need to measure the power used on dozens and dozens of servers. I need a breakdown by server, not a total for the room. (more specifically, I want to calculate the power cost per server per year). Added difficulty: I want to do this non-intrusively. I can't go round, yanking out cables and inserting in-line meters. So, I got a clamp meter. But contrary to my expectation, that only works if I can separate the individual wires in the power cable - not an option. So, I'm stuck. Is there any non-intrusive way to measure the power consumption of my servers? Andrew
Re: Measuring power
2009/4/29 Andrew Beattie and...@tug.com: This is a long way off topic but I'm hoping that someone here might be able to share a clue... I need to measure the power used on dozens and dozens of servers. I need a breakdown by server, not a total for the room. (more specifically, I want to calculate the power cost per server per year). Added difficulty: I want to do this non-intrusively. I can't go round, yanking out cables and inserting in-line meters. So, I got a clamp meter. But contrary to my expectation, that only works if I can separate the individual wires in the power cable - not an option. So, I'm stuck. Is there any non-intrusive way to measure the power consumption of my servers? Andrew Do you need exact figures? And, by non-intrusively, do you mean from a software point of view as well? I'm thinking a utility like powertop could be of use, although I'm not sure how well that would compare with losses in the PSU and fans and suchlike. Presumably you could use an in-line meter for one server, and use this to work out a best guess for others. Adrian.
Re: Measuring power
I'd second the option of getting a similar machine and comparing that. The power usage doesn't vary much more than about 20% against CPU load in my experience, unfortunately. If you're feeling brave you can separate the live wire from the rest _reasonably_ safely. The trick is using a very sharp knife and bending the cable and nicking the outer sheath at the outside 'edge' of the curve. What tends to happen is the sheath tears and with successive small slices will tear more reveal enough for you to work the rest of the sheath off safely. Rubber gloves mats too, obviously :-) Paul On Wed, Apr 29, 2009 at 9:42 AM, Andrew Beattie and...@tug.com wrote: This is a long way off topic but I'm hoping that someone here might be able to share a clue... I need to measure the power used on dozens and dozens of servers. I need a breakdown by server, not a total for the room. (more specifically, I want to calculate the power cost per server per year). Added difficulty: I want to do this non-intrusively. I can't go round, yanking out cables and inserting in-line meters. So, I got a clamp meter. But contrary to my expectation, that only works if I can separate the individual wires in the power cable - not an option. So, I'm stuck. Is there any non-intrusive way to measure the power consumption of my servers? Andrew
Re: Measuring power
On Wed, Apr 29, 2009 at 10:05:01AM +0100, Michael Lush wrote: A frend of mine has a Power Owl http://www.greenstamp.co.uk/product_info.php/cPath/33/products_id/68?gclid=COmqvdzclZoCFR4hnAodx3XzNw You clip the sensor round the mains power were it comes in to the electricity meter. I guess you could clip it round the server power cable I'd be inclined to calibrate it against a inline power meter before use... I have a similar meter from Efergy at home and once spent a frustrating weekend trying to get it to agree with the official meter. Eventually I realised that these meters are measuring VA, not W. The same for resistive loads, but considerably different for inductive loads such as CFL and electric motors. I have one 32W CFL that draws 67VA. Of course VA maybe what you want. I got a few inline power meters from Maplins when they were on offer at 9.99 http://www.maplin.co.uk/Module.aspx?ModuleNo=38343 (But they're now 22.99!) Ah, the current deal is http://www.maplin.co.uk/Module.aspx?ModuleNo=223573 @ 7.99. OK for domestic use but that doesn't really help in a rack or if you can't unplug the server. There must be a market for better power monitoring: I'd like a meter that piggy-backs in the circuit-breaker socket of a distribution board so that I can monitor each circuit. Individual devices would be even better! -- Chris Benson
Re: Measuring power
2009/4/29 Paul Makepeace pa...@paulm.com: If you're feeling brave you can separate the live wire from the rest _reasonably_ safely. The trick is using a very sharp knife and bending the cable and nicking the outer sheath at the outside 'edge' of the curve. What tends to happen is the sheath tears and with successive small slices will tear more reveal enough for you to work the rest of the sheath off safely. Rubber gloves mats too, obviously :-) Oh, go on, use your teeth, and stand in a bucket of water. People wonder how anyone with an entry in the Darwin awards could be so stupid. -- Jasper
Re: Measuring power
On Wed, Apr 29, 2009 at 10:56 AM, Jasper jaspermcc...@gmail.com wrote: 2009/4/29 Paul Makepeace pa...@paulm.com: If you're feeling brave you can separate the live wire from the rest _reasonably_ safely. The trick is using a very sharp knife and bending the cable and nicking the outer sheath at the outside 'edge' of the curve. What tends to happen is the sheath tears and with successive small slices will tear more reveal enough for you to work the rest of the sheath off safely. Rubber gloves mats too, obviously :-) Oh, go on, use your teeth, and stand in a bucket of water. People wonder how anyone with an entry in the Darwin awards could be so stupid. More stupid than running a server farm where individual machines can't be switched off? ;-)
Re: Measuring power
On Wed, 2009-04-29 at 09:42 +0100, Andrew Beattie wrote: This is a long way off topic but I'm hoping that someone here might be able to share a clue... I need to measure the power used on dozens and dozens of servers. I need a breakdown by server, not a total for the room. (more specifically, I want to calculate the power cost per server per year). Added difficulty: I want to do this non-intrusively. I can't go round, yanking out cables and inserting in-line meters. So, I got a clamp meter. But contrary to my expectation, that only works if I can separate the individual wires in the power cable - not an option. So, I'm stuck. Is there any non-intrusive way to measure the power consumption of my servers? No easy way, it seems. If you *really* need to do this to the point of entertaining loony schemes... One possibility might be to measure the heat output of the servers, rather than the electricity input. If you assume that the servers lose heat mainly by pumping hot air out with fans, you'd need to measure both the air flow rate and the difference between output air temperature and input air temperature.
Re: Measuring power
2009/4/29 Nick Cleaton n...@cleaton.net: One possibility might be to measure the heat output of the servers, rather than the electricity input. If you assume that the servers lose heat mainly by pumping hot air out with fans, you'd need to measure both the air flow rate and the difference between output air temperature and input air temperature. Fascinating, but there would be no way of calibrating this, unless you built an identical server room that you could play with. The processor temperatures (with some calibration to room temperature) might have a mappable relation to power consumption. -- Jasper
Re: Measuring power
On Wed, 2009-04-29 at 12:23 +0100, Jasper wrote: 2009/4/29 Nick Cleaton n...@cleaton.net: One possibility might be to measure the heat output of the servers, rather than the electricity input. If you assume that the servers lose heat mainly by pumping hot air out with fans, you'd need to measure both the air flow rate and the difference between output air temperature and input air temperature. Fascinating, but there would be no way of calibrating this, unless you built an identical server room that you could play with. Not really. If you know how much air is being heated up by how many degrees each second, you can look up the specific heat capacity of air and convert that number to the power in Watts required to perform that heating. A computer is just a complicated heater, after all. Some energy will escape as RF radiation instead of heat, but server cases are designed to block that stuff, so probably not much. I think the main error would be due to heat radiated from the server case or conducted to the rack or the adjacent servers. I'd guess (based on nothing in particular) that most of the heat would come out via the fans though.
Re: Measuring power
On Wed, Apr 29, 2009 at 09:42:55AM +0100, Andrew Beattie wrote: I can't go round, yanking out cables and inserting in-line meters. So, I got a clamp meter. But contrary to my expectation, that only works if I can separate the individual wires in the power cable - not an option. So, I'm stuck. Is there any non-intrusive way to measure the power consumption of my servers? You could measure the heat output from each server, as, averaged over a reasonably long period, that is == energy consumption. However, to do that accurately would be even more disruptive than just scheduling some downtime to use an inline meter. Or you could yank the power and see how long the UPSes last :-) -- David Cantrell | Nth greatest programmer in the world Did you know that shotguns taste like candy canes? Put the barrel in your mouth and pull the trigger for an extra blast of minty goodness!
Re: Measuring power
2009/4/29 Nick Cleaton n...@cleaton.net: On Wed, 2009-04-29 at 12:23 +0100, Jasper wrote: 2009/4/29 Nick Cleaton n...@cleaton.net: One possibility might be to measure the heat output of the servers, rather than the electricity input. If you assume that the servers lose heat mainly by pumping hot air out with fans, you'd need to measure both the air flow rate and the difference between output air temperature and input air temperature. Fascinating, but there would be no way of calibrating this, unless you built an identical server room that you could play with. Not really. If you know how much air is being heated up by how many degrees each second, you can look up the specific heat capacity of air and convert that number to the power in Watts required to perform that heating. My point was that the room would not be a closed system.. -- Jasper
Re: Measuring power
On Wed, Apr 29, 2009 at 10:53:30AM +0100, Chris Benson wrote: I got a few inline power meters from Maplins when they were on offer at 9.99 http://www.maplin.co.uk/Module.aspx?ModuleNo=38343 (But they're now 22.99!) Ah, the current deal is http://www.maplin.co.uk/Module.aspx?ModuleNo=223573 @ 7.99. OK for domestic use but that doesn't really help in a rack or if you can't unplug the server. The price is right, come on down! I went to Maplins and bought one at lunchtime. It's quite big (too wide to really plug in anything next to it), but seems to measure things OK (well, it produced plausible numbers). I've only tried a flat panel and a phone charger so far, because the office manager was in the kitchen. Nicholas Clark
