From dpelleg+ at cs.cmu.edu Mon Nov 8 14:20:49 2004 From: dpelleg+ at cs.cmu.edu (Dan Pelleg) Date: Mon, 8 Nov 2004 14:20:49 -0500 Subject: [auton-users] RE: cycle servers Message-ID: <16783.50961.859335.152795@lark.auton.cs.cmu.edu> Just an update on the usage of cycle servers. The deadline of the project I'm working on is not over yet, and I still need large amounts of processing power before it is. I can imagine how hard this is on others who need the servers, and I am grateful for everyone's cooperation on this. I will try to keep jobs off lop1 and lop2, lira and lina, to give you a more predicable usage pattern. You can also use other machines that are otherwise idle. Note that I do not expect any other machines to become available in the next 48 hours. Please, if you run jobs, run all of them under nice(1), and if they have already started, renice them. This will have no effect unless other jobs are running on the same machine, in which case they will get a lower priority. Andrew's original message (of 10/21) follows. > Over the next few weeks please avoid using the cycle servers for big > jobs. Several sponsored research projects need results in the short term > and we want to use the machines for those. We'll let youknow when this > period ends: probably mid-November. If a machine looks empty and has > been unused for a while then feel free to run your code provided it > doesn't use a whole lot of memory, but in "nice" mode. > > Thanks, > > Andrew --Dan From awm at cs.cmu.edu Mon Nov 8 15:09:08 2004 From: awm at cs.cmu.edu (Andrew W Moore) Date: Mon, 08 Nov 2004 15:09:08 -0500 Subject: [auton-users] Re: cycle servers Message-ID: <418FD264.4080806@cs.cmu.edu> Let me echo Dan in thanking you guys, and let me comment that this is indeed a critical need not just a typical conference deadline crunch. This is for results for sponsors that fund a large fraction of our past present and hopefully future work, so Dan, Rahul, Jennifer and others doing this heroic work are helping keeping us all off the streets and out of the gutter in doing it. Plus much of the money for the cycle servers came from these sponsors. Andrew From dpelleg+ at cs.cmu.edu Thu Nov 18 08:53:05 2004 From: dpelleg+ at cs.cmu.edu (Dan Pelleg) Date: Thu, 18 Nov 2004 08:53:05 -0500 Subject: [auton-users] cycle servers free again Message-ID: <16796.43329.104995.564924@lark.auton.cs.cmu.edu> My project deadline is now officially over, and the compute servers are back to normal usage rules. A big thank you to everyone for your cooperation during this period! -- Dan Pelleg From komarek at cmu.edu Thu Nov 18 11:55:39 2004 From: komarek at cmu.edu (Paul Komarek) Date: Thu, 18 Nov 2004 11:55:39 -0500 Subject: [auton-users] Reservation request Message-ID: <419CD40B.9070806@cmu.edu> Hello, I would like to reserve a 4GB lop for 1 week. I need to repeat SVM timing experiments for several parameter configs, and we all know how slow SVM is. ;-) How about lop7 until the 25th of Nov? -Paul Komarek From komarek at cmu.edu Fri Nov 19 10:53:11 2004 From: komarek at cmu.edu (Paul Komarek) Date: Fri, 19 Nov 2004 10:53:11 -0500 Subject: [auton-users] Reservation request In-Reply-To: <419CD40B.9070806@cmu.edu> References: <419CD40B.9070806@cmu.edu> Message-ID: <419E16E7.5070902@cmu.edu> I received only one reply, and it seems it is best if I reserve loq1. So I'm reserving loq1 through next Friday. -Paul Komarek Paul Komarek wrote: > Hello, > > I would like to reserve a 4GB lop for 1 week. I need to repeat SVM > timing experiments for several parameter configs, and we all know how > slow SVM is. ;-) > > How about lop7 until the 25th of Nov? > > -Paul Komarek > From jmjoseph at andrew.cmu.edu Mon Nov 29 12:25:13 2004 From: jmjoseph at andrew.cmu.edu (Jacob Joseph) Date: Mon, 29 Nov 2004 12:25:13 -0500 Subject: [auton-users] Lop2, q1, q3 Reservation Message-ID: <41AB5B79.4000009@andrew.cmu.edu> Hi all. In short, I have reserved Lop2, Loq1, and Loq3 for the Condor job batching system. The full explanation follows. To ease the need for the current reservation system and to more reasonably balance load at times of high demand, we have been testing the Condor batch submission system. We've reached a point where we are ready to test with a more realistic environment. The goal at this point is to discover any shortcomings in the queuing system and fine tune the user experience. Once the bugs are worked out, we should like to remove direct access to many of the cluster machines, leaving only a few machines with interactive access for testing and submission. Any machine would of course still be available for reservation on a case-by-case basis. Reducing the number of 'front-end' machines should greatly reduce the complexity currently present in selecting an available machine while the batching system would ensure fair usage. If you've got 10 minutes to spare and have some jobs to run, I certainly am looking for volunteers to help squash out the bugs. If you've used condor before, use lop1 for submission. Rather than including all of the submission details here, I ask that you do send me an email if you're interested. -Jacob