dropping dual clients

[FLECOM]

well ever since i got my mega borg i have been pushing two standard clients to each machine, since they are all dual core machines, i ran each client at "90%"... but they are still getting too hot, and its been an issue... first the AC is off at night so the machines get extremely hot, causing a huge amount of hard-disk failures (ive had about ~40 drives die in 2 months - yay dell warranty) also they have been making the rooms unbearably hot and the fans get really loud...

so the solution, i am only putting a single client at "100%" and seeing how that goes... hopefully that will solve the problem... after that i only have two options, turn the machines off at night, or remove F@H from them all together...

i am sad

 

[Celerator]

Too bad, but 100% single cores are better than nothing. Let's hope the system can dissapate the heat.

 

[MrGuvernment]

if the computers are killing HD's that often why dont you move them? or find a better solution...

 

[FLECOM]

well they are not my computers, they are workstations and the A/C apparently cant keep up with the extra heat load, plus the AC is off at night so they are getting very very hot and the cases are not very well designed, so the drives die...

i cant really come up with a better solution quite frankly

 

[Jon855]

I guess when it all comes down to it, make sure the workstations are running because that's the main thing that's important -> work rigs needs to be functional...

Whatever you decide to do Flecom, you can rest assure you'll have our respect no matter how bad it could get... you are untouchable.

 

[gnewbury]

There seems to be a couple of solutions.

If heat is a problem at night, but not during the day, run one instance as a scheduled task that cuts off at night.

Write to RAM and let those drives sleep.

I realize that for your vast fleet changes will be difficult,

 

[marty9876]

I'd suggest taking some time and fining the "thermal envelope" so to speak.

My experience is that it's a non-linear relationship between % of processing and heat output.

IE running @ 25% produces a CPU temp of 50 C and 10 BTU's (making these numbers up) [having the computer on to begin with is 46C CPU and 8 BTU's at idle]

@ 50 % CPU temps of 52 C and making 12 BTU's
@ 75% CPU temps of 53 C and making 14 BTU's
@ 85% CPU temps of 57C and making 22 BTU's
@ 95% CPU tamps of 60C and making 30 BTU's
@ 100% CPU temps of 63C and making 45 BTU's

ETC. Same holds true for the electrical draw which I image is where this all starts from.

(I made up the BTU's totally, too early for me to estimate BTU output)

Point being is the boxes might make 80% less heat if ran at 60% CPU usage across both cores where as might only make 50% less heat if one core is maxed out.

On a single CPU box this is pretty easy, on a dual core box there is just more variables. Maybe the easiest thing would be to grab a Kill-a-watt and record/graph the watt draw in relationship to the CPU usage (under signal and dual client configs with different CPU usage). Then for the points part... compare that against estimated PPD of the clients for the net loss.

Grab a box and start working with it perhaps.

No borg = 100% net loss so...

 

[MN Scout]

I like the idea of having one be a scheduled task that turns off at night.

 

[BillR]

Damnit, I hate it when Marty makes sense, thank God it doesn't happen too often

 

[BillR]

well they are not my computers, they are workstations and the A/C apparently cant keep up with the extra heat load, plus the AC is off at night so they are getting very very hot and the cases are not very well designed, so the drives die...

i cant really come up with a better solution quite frankly

The reality is, if you turn the machines off at night which I assume was their usual practice the office will have time to cool down, less gear will break and the boss will be less inclined to find a new IT manager who plays childrens's games on an old G-2

In this case the single units make more sense because of the longer deadlines and you will still have really good output. Just no sense risking the nicest borg ever man.

 

[HAF72]

did I just heard flecom is having trouble folding? woooooohooooooooooooooo! well.. if he has to stop it will still take years and years and years and years to catch up

 

[FLECOM]

The reality is, if you turn the machines off at night which I assume was their usual practice the office will have time to cool down, less gear will break and the boss will be less inclined to find a new IT manager who plays childrens's games on an old G-2

In this case the single units make more sense because of the longer deadlines and you will still have really good output. Just no sense risking the nicest borg ever man.

no usually machines stay on at night... this might change i dont know yet...

i think i need to buy a kill-a-watt

 

[marty9876]

One thing I forgot thinking about this, are you running big packets on these? I think I recall that you are.

Big units lay into the SSE stuff much, much more than other units (Tets/Ribo's). More transistors that are firing, more heat is produced.

I can honestly tell if one of my boxes is running a big packet WU or now just by placing my had on top of the case and feeling the exterior case temperature. It's a very noticeable difference (basically warm vs. not warm) between normal WU's and big WU's.

 

[FLECOM]

One thing I forgot thinking about this, are you running big packets on these? I think I recall that you are.

Big units lay into the SSE stuff much, much more than other units (Tets/Ribo's). More transistors that are firing, more heat is produced.

I can honestly tell if one of my boxes is running a big packet WU or now just by placing my had on top of the case and feeling the exterior case temperature. It's a very noticeable difference (basically warm vs. not warm) between normal WU's and big WU's.

yes i do have big packets turned on, i will look into that