 |
 |
On Mon, Oct 03, 2005 at 11:43:45AM +0200 or thereabouts, Antonio Delgado Peris wrote: > To be sure, I was just referring to the case where you scale the nodes (as > Steve mentioned), so you are sure anyway that the CPU and WClock limits > are long enough for your jobs, no matter what gets published. > > In that case, I guess the average value being published gives a more > realistic idea of the speed of your nodes. > > In case you don't scale, then of course you should always publish the > worst case. Yes in principal it does not matter what you scale to however picking something sensible like the average of your nodes with some future proofing. We were until two weeks ago normalising to 211 spec int for historical reasons and scaling factors varied from 3 to 5. We have now moved to 1000 spec int which makes everything closer to 1. Steve > > Antonio. > > > > On Mon, 3 Oct 2005, Louis Poncet wrote: > > > So the problematic is do people prefer "worst case" or "average". > > > > Personaly i think the worst case is safer. > > > > Lp > > -- > > Laird Louis Poncet > > Where: Bat28-R-003 CERN > > CH-1211 Geneve 23 > > Mail : [log in to unmask] > > Phone: +41(0)227.674.231 > > LAL / IN2P3 / CNRS / CERN > > Problem >> RTFM then google it ! > > > > > > Le 3 oct. 05 à 11:37, Antonio Delgado Peris a écrit : > > > > > On Wed, 28 Sep 2005, Steve Traylen wrote: > > > > > > > > > > > Or you can scale all of your nodes to the same value of your choice > > > > and publish that. > > > > > > > > > > > > > And it wouldn't be a bad idea that the chosen value was something like the > > > average of your nodes, so users could not only be sure that the max time > > > limits are long enough (for that any value would do), but also might > > > estimate (on average) how long their jobs will really take. > > > > > > Antonio. > > > > > > > > > > > > > > > > > > > > See > > > > > > > > cputmult and wallmult configuration values at > > > > > > > > http://www.clusterresources.com/products/torque/docs/1.1momconfig.shtml > > > > > > > > Steve > > > > > > > > > > > > > > > > > > -- > > > > > > > > > > > > _______________________________________ > > > Antonio Delgado Peris > > > _______________________________________ > > > IT Division - GD Group - EIS section > > > CERN. CH-1211 Genève 23 (Switzerland) > > > Office: 28-R-017 > > > Tel.: +41 22 76 72227 > > > Email: [log in to unmask] > > > _______________________________________ > > > > > > > > > -- > > > > _______________________________________ > Antonio Delgado Peris > _______________________________________ > IT Division - GD Group - EIS section > CERN. CH-1211 Genève 23 (Switzerland) > Office: 28-R-017 > Tel.: +41 22 76 72227 > Email: [log in to unmask] > _______________________________________ -- Steve Traylen [log in to unmask] http://www.gridpp.ac.uk/