Quantcast
Channel: StudioSysAdmins Message Board
Viewing all articles
Browse latest Browse all 3749

Number of blades per circuit 30 amp/220 watt circuit

$
0
0
By Klaus Steden -

I've got a Dell R620 coming online and the IPMI controller has that kind of info:

-- cut --
Fan1A RPM        | 3960 RPM          | ok
Fan2A RPM        | 3840 RPM          | ok
Fan3A RPM        | 3840 RPM          | ok
Fan4A RPM        | 3840 RPM          | ok
Fan5A RPM        | 3840 RPM          | ok
Fan6A RPM        | 3840 RPM          | ok
Inlet Temp       | 20 degrees C      | ok
Exhaust Temp     | 36 degrees C      | ok
Temp             | disabled          | ns
Temp             | disabled          | ns
Current 1        | disabled          | ns
Current 2        | 1 Amps            | ok
Voltage 1        | disabled          | ns
Voltage 2        | 118 Volts         | ok
Pwr Consumption  | 112 Watts         | ok
Fan7A RPM        | 3960 RPM          | ok
Fan1B RPM        | 3600 RPM          | ok
Fan2B RPM        | 3720 RPM          | ok
Fan3B RPM        | 3720 RPM          | ok
Fan4B RPM        | 3720 RPM          | ok
Fan5B RPM        | 3720 RPM          | ok
Fan6B RPM        | 3720 RPM          | ok
Fan7B RPM        | 3720 RPM          | ok
-- cut --

Currently running on a single PSU hence the lack of current and voltage readings for PDU 1. That's the kind of thing I want to see vendors implement across the board -- the IPMI standard is great, but it would be even better if they were able to standardize on what quantities were presented.

cheers,
Klaus


From: greg whynott [greg.whynott@gmail.com]
Sent: Thursday, February 28, 2013 10:04 AM
To: discuss@studiosysadmins.com
Subject: Re: [SSA-Discuss] Number of blades per circuit 30 amp/220 watt circuit

most systems will also allow you to query the PSU to see what its up to now a days don't they?



On Thu, Feb 28, 2013 at 12:26 PM, Nick Allevato <nick@weareroyale.com> wrote:
Empirical measurement is the way to go.

Grab some metered PDU's from Sergio @ CRE and add nodes to each circuit during rendering until you get within a node's worth of amps on that particular circuit and you should be gooooood to go.

As for planning for max PSU rating...generally nodes eat up a good amount of power...not max, sure, but dual 1U render nodes will eat up a pretty good amount of power.

-nick



Nick Allevato | Information Technology | 
 

On Feb 28, 2013, at 7:49 AM, greg whynott wrote:

urandom actually...

for a in 0 1 2 3 4 5 6 7 8 9 10;do cat /dev/urandom > /dev/null & done

where the count is the number of cores you have. 

there is a package you can install called cpuburn which includes the burn* bins for various cpu types.   probably does a better job at 'filling' the cpu.

-g



On Thu, Feb 28, 2013 at 8:53 AM, Greg Whynott <greg.whynott@gmail.com> wrote:
Cat Dev random works too.

G

Sent while driving.

On 2013-02-27, at 23:23, Greg Ercolano <erco_mlist@seriss.com> wrote:

> On 02/27/13 19:16, bill@yuco.com wrote:
>> Also plan on getting a metered 220 power strip to see the actual draw.
>
>    It's been a long time since I wrangled render farms, but..
>
>    +1 on empirical checking of current draw during heavy use, then pad
>    that as needed. So if it runs at 300W, then perhaps plan for 500W
>    per node, or some such.
>
>    But I wouldn't plan based on the max rating of the power supply;
>    that's surely overkill, as that's the rating the supply will max out,
>    not its normal current draw.
>
>    FWIW if you want to slam the box, I found this one liner is a good way
>    to just slam a cpu to 100%; run as many of these as you have cpus
>    to max them all:
>
> perl -e '$t=0; while(1) { $a = sin($t); $t+=.1; }'
>
>    See *warning* below; I melted a 2 cpu laptop once by running two instances
>    of that one liner..!
>
>    This will heavily exercise the cpu's floating point math, getting it
>    hot and active with the fans and such, but does no memory access,
>    and does not exercise the disks.
>
>    You may also want to exercise the disk(s), as they eat power and generate heat
>    too. Running this will keep your local drive very busy:
>
> find / -mount > /dev/null
>
>    On windows (not in front of it now), I guess something similar might be:
>
> DIR /S C:\ > NUL
>
>    Also, of course run heavy renders (exercise floating cpu use) and comps
>    (exercises cpu+I/O and disk) just to see if they magically make the box
>    use more power than the above.
>
>
> *WARNING*
>    Be /very/ careful with that one liner perl command.
>
>    I overheated a 2 cpu laptop once (HP Pavilion dv2415nr running linux) by running
>    two instances of that one liner in two terminal windows. It did an auto-thermal
>    shutdown after about 3 or 4 minutes, but the box would also not come up again
>    even hours later. (Next day it kinda worked again, but was always flaky from then on..)
>
>    Granted laptops are much weaker than render nodes, but I was still surprised
>    at the behavior.
>
>    Anyway, keep your eye on heat.
> To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe

To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


Viewing all articles
Browse latest Browse all 3749

Trending Articles