Yikes, I won't do that again!

Nothing But Idle Time
Nothing But Idl...
Joined: 24 Aug 05
Posts: 158
Credit: 289204
RAC: 0
Topic 193050

I always run Einstein and Rosetta 50/50 on my P4/HT machine. My current data pack has been running tasks of 51-52 hours in length. Well, some poor guy's/gal's wingman aborted a task and it was sent to me. So I thought I would help him/her out by running the new task in addition to the Einstein task I was already running. Yikes, the completion time jumped from 51 hours to 66 hours on both tasks. From now on my wingman (wingperson?)can wait in line just like every other wingman.

testr2
testr2
Joined: 18 Jan 05
Posts: 13
Credit: 653354
RAC: 0

Yikes, I won't do that again!

"I always run Einstein and Rosetta 50/50 on my P4/HT machine. My current data pack has been running tasks of 51-52 hours in length. Well, some poor guy's/gal's wingman aborted a task and it was sent to me. So I thought I would help him/her out by running the new task in addition to the Einstein task I was already running. Yikes, the completion time jumped from 51 hours to 66 hours on both tasks. From now on my wingman (wingperson?)can wait in line just like every other wingman."

I was around in the 70's and 80's and saw some perfectly good computer people make the same mistake, sometimes with far worse results than you got.

Udo
Udo
Joined: 19 May 05
Posts: 203
Credit: 8945570
RAC: 0

RE: Yikes, the completion

Quote:
Yikes, the completion time jumped from 51 hours to 66 hours on both tasks.

...thats the effect of the Intel hyperthreading architecture...
2 different tasks on a single core (but hyperthreading) CPU is generally better (more performant) than 2 times the same tasks...
...struggling for CPU core ressources which are virtually 2 times inside the CPU but in reality only once!

Udo

th3
th3
Joined: 24 Aug 06
Posts: 208
Credit: 2208434
RAC: 0

Yeah, HT can be nice, i would

Yeah, HT can be nice, i would think its better to finish 2 WUs in 66hours instead of 1 WU in 50 hours?

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686138939
RAC: 552243

RE: Yeah, HT can be nice, i

Message 71306 in response to message 71305

Quote:
Yeah, HT can be nice, i would think its better to finish 2 WUs in 66hours instead of 1 WU in 50 hours?

The 50 hours comparision figure was also hyperthreaded, if I understand this correctly, but with a different mix (Einstein/Rosetta vs Einstein/Einstein). If run w/o Hyperthreading, an Einstein WU of the size in question should complete in far less than 50 hrs on this PC, but almost certainly not as fast as 25 hours (but 33 hrs could be possible). So basically HT might be no big advantage if you are just crunching Einstein, but does work in your favor if you are running Einstein along with (say) Rosetta.

Interesting stuff, thanks for posting it here.

CU

BRM

Nothing But Idle Time
Nothing But Idl...
Joined: 24 Aug 05
Posts: 158
Credit: 289204
RAC: 0

RE: RE: Yeah, HT can be

Message 71307 in response to message 71306

Quote:
Quote:
Yeah, HT can be nice, i would think its better to finish 2 WUs in 66hours instead of 1 WU in 50 hours?

The 50 hours comparision figure was also hyperthreaded, if I understand this correctly, but with a different mix (Einstein/Rosetta vs Einstein/Einstein). If run w/o Hyperthreading, an Einstein WU of the size in question should complete in far less than 50 hrs on this PC, but almost certainly not as fast as 25 hours (but 33 hrs could be possible). So basically HT might be no big advantage if you are just crunching Einstein, but does work in your favor if you are running Einstein along with (say) Rosetta.

Interesting stuff, thanks for posting it here.

CU

BRM

Your interpretation is correct. I've known about the Einstein/Rosetta vs Einstein/Einstein scenario since I joined in 2005. It's noteworthy that the jump from 51 to 66 hours happened after the first Einstein/Rosetta units crunched for a day and then the second Einstein task was started. So had I started the 2 Einstein tasks simultaneously the final completion time for both would probably have been longer than 66 hours. With Akos' S4 apps the time differences were much smaller than with the less optimized S5 app.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.