Cooler servers for the Internet
IBM is announcing several new servers today. One interesting feature of the press release is a new focus on power consumption and cooling. IBM directly targets the high heat generated by the SUN servers. Compared to a Sun Server, the new IBM servers have 1/3 the heat dissipation, based on the published specifications. IBM eServer p660 is 3,294 BTUs/hr compared with 9,420 BTUs/hr for a Sun Fire 3800. Of course, these numbers are pretty bogus. Most computer manufacture specification sheets are useless for accurately forecasting the power consumption of the equipment. I suspect, after Sun's PR people realize they are getting beat up, they will go back and better calculate the power consumption figures for their servers. I'm not very interested in who really has the coolest, most efficient computers. I am interested in getting accurate information for planning purposes. If this leads to computer vendors publishing more accurate information, great. I'm afraid instead, the pendulum will swing the other direction and vendors will begin understating their true power requirements.
From: Sean Donelan <sean@donelan.com>
I'm not very interested in who really has the coolest, most efficient computers. I am interested in getting accurate information for planning purposes. If this leads to computer vendors publishing more accurate information, great. I'm afraid instead, the pendulum will swing the other direction and vendors will begin understating their true power requirements.
At least one other manufacturer does want to compete on wall-plug efficiency: http://www.rlxtechnologies.com/ Chris
Sean, I could not be in more agreement with you. I have written a white paper that discusses this issue. It covers some of the basics of cooling and power but it also takes it down to Ohms Law and shows a few excel worksheets for calculations of single systems as well as roll-up's for racks of equipment. Here is a link to the white-paper: http://www.colosource.com/whitepaper/Coolingandpowerpaper.pdf Please let me know what you think if you have a chance to give it a read. Thanks, Kevin Facinelli Director of Operations Crystal Group Inc www.crystalpc.com 319-378-1636 x227 --- Sean Donelan <sean@donelan.com> wrote:
IBM is announcing several new servers today. One interesting feature of the press release is a new focus on power consumption and cooling.
IBM directly targets the high heat generated by the SUN servers. Compared to a Sun Server, the new IBM servers have 1/3 the heat dissipation, based on the published specifications. IBM eServer p660 is 3,294 BTUs/hr compared with 9,420 BTUs/hr for a Sun Fire 3800.
Of course, these numbers are pretty bogus. Most computer manufacture specification sheets are useless for accurately forecasting the power consumption of the equipment. I suspect, after Sun's PR people realize they are getting beat up, they will go back and better calculate the power consumption figures for their servers.
I'm not very interested in who really has the coolest, most efficient computers. I am interested in getting accurate information for planning purposes. If this leads to computer vendors publishing more accurate information, great. I'm afraid instead, the pendulum will swing the other direction and vendors will begin understating their true power requirements.
===== Kevin Facinelli www.colosource.com webmaster@colosource.com __________________________________________________ Do You Yahoo!? Yahoo! Auctions - buy the things you want at great prices http://auctions.yahoo.com/
Unnamed Administration sources reported that Sean Donelan said:
Of course, these numbers are pretty bogus. Most computer manufacture specification sheets are useless for accurately forecasting the power consumption of the equipment. I suspect, after Sun's PR people realize they are getting beat up, they will go back and better calculate the power consumption figures for their servers.
I'm not very interested in who really has the coolest, most efficient computers. I am interested in getting accurate information for planning purposes. If this leads to computer vendors publishing more accurate information, great. I'm afraid instead, the pendulum will swing the other direction and vendors will begin understating their true power requirements.
The best test I can come up with at the drop of a hat: Plug server into std. watt-hour meter. Run benchmark X, Y and/or Z for one hour. Note KWH's consumed. This is so trivial a software house could do it if they had an electrician buy the meter and wire it up with plugs. It will tell you true KWH's, as that's what they measure. The meters are highly tracable back to NIST/etc standards [1] and they are cheap. Radical idea: why not call The Donelan Test, and demand it from your vendors? 1] Want stds that have been beaten on? Use those proven by products with lots of money changing hands. -- A host is a host from coast to coast.................wb8foz@nrk.com & no one will talk to a host that's close........[v].(301) 56-LINUX Unless the host (that isn't close).........................pob 1433 is busy, hung or dead....................................20915-1433
David, Many times early in the development process hardware options are being evaluated prior to purchase....The need for accurate power information extends beyond system deployment, it starts during systems specification. I have come across many situations where companies were not making consideration for power available by their power supply. The thought was it has the slots so stick the extra cards in (very prevent). If a system is over loading a power supply it is very likely to have premature failure along with potential to damage to system components. I do like your idea of average power requirement metering using simple test equipment when possible. Kevin Kevin Facinelli Director of Operations Crystal Group Inc www.crystalpc.com 319-378-1636 x227 --- David Lesher <wb8foz@nrk.com> wrote:
Unnamed Administration sources reported that Sean Donelan said:
Of course, these numbers are pretty bogus. Most
specification sheets are useless for accurately forecasting the power consumption of the equipment. I suspect, after Sun's PR people realize they are getting beat up, they will go back and better calculate the power consumption figures for their servers.
I'm not very interested in who really has the coolest, most efficient computers. I am interested in getting accurate information for planning purposes. If this leads to computer vendors
information, great. I'm afraid instead, the
computer manufacture publishing more accurate pendulum will swing the
other direction and vendors will begin understating their true power requirements.
The best test I can come up with at the drop of a hat:
Plug server into std. watt-hour meter.
Run benchmark X, Y and/or Z for one hour.
Note KWH's consumed.
This is so trivial a software house could do it if they had an electrician buy the meter and wire it up with plugs.
It will tell you true KWH's, as that's what they measure. The meters are highly tracable back to NIST/etc standards [1] and they are cheap.
Radical idea: why not call The Donelan Test, and demand it from your vendors?
1] Want stds that have been beaten on? Use those proven by products with lots of money changing hands.
-- A host is a host from coast to coast.................wb8foz@nrk.com & no one will talk to a host that's close........[v].(301) 56-LINUX Unless the host (that isn't close).........................pob 1433 is busy, hung or dead....................................20915-1433
===== Kevin Facinelli www.colosource.com webmaster@colosource.com __________________________________________________ Do You Yahoo!? Yahoo! Auctions - buy the things you want at great prices http://auctions.yahoo.com/
On Mon, 23 Apr 2001 05:39:01 PDT, Sean Donelan <sean@donelan.com> said:
I'm not very interested in who really has the coolest, most efficient computers. I am interested in getting accurate information for planning purposes. If this leads to computer vendors publishing more accurate information, great. I'm afraid instead, the pendulum will swing the other direction and vendors will begin understating their true power requirements.
I don't have numbers handy for the IBM boxes, but note that the p680 can go 6 to 24 CPU, and from 4G to 96G of RAM. Hardly a case where one number will fit all. IBM at least *used* to provide power info on a per-component basis, but you had to get the 'Site Planning and Preparation' manual for the system in question. I also know that SGI has information on amperage considerations for their O2K, Onyx and Onyx2 rack-mount servers, where you *can* get into the situation where you have open slots, but be unable to put a high-draw card in because the power supply for the cage isn't rated for it. The only way to get "more accurate" info is to line-item each component and add up the numbers for *your* configuration. -- Valdis Kletnieks Operating Systems Analyst Virginia Tech
participants (5)
-
chrisb@kippona.com
-
David Lesher
-
Kevin Facinelli
-
Sean Donelan
-
Valdis.Kletnieks@vt.edu