Mailing List Archives
Public Access
|
|
|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[HTCondor-users] WG: Users priorities problem
- Date: Wed, 31 Jul 2013 14:39:13 +0100 (BST)
- From: daniel popu <dpopu@xxxxxxxxx>
- Subject: [HTCondor-users] WG: Users priorities problem
Hello,
I sent the mail below and I got the answer on
https://www-auth.cs.wisc.edu/lists/htcondor-users/2013-July/msg00118.shtml
How could I continue that thread of discussion?
First of all I wish to thank Steffen Grunewald for his answer.
Second I wish to ask other items.
1) Steffen mention that I should take care of CLAIM_WORKLIFE to not put it too small.
For the moment I used
UWCS_CLAIM_WORKLIFE = (3 * $(MINUTE))
3 minutes is too short?
How can I detect if my throughput is harmed because of too short lease time?
2) Resources usage Percentage
In my pool I found the next situation:
There are 150 free cores.
We have only one active user.
He starts a simulation but he gets maximum 90 cores.
With other occasion he could get maximum of 140 cores.(so there is no constraint from OS).
Is there a parameter which influences the maximum number of cores allocated to a user?
All the best,
Daniel
----- Weitergeleitete Message -----
Von: daniel popu <dpopu@xxxxxxxxx>
An: "htcondor-users@xxxxxxxxxxx" <htcondor-users@xxxxxxxxxxx>
CC:
Gesendet: 11:36 Dienstag, 30.Juli 2013
Betreff: Users priorities problem
Hello,
I build a pool of computers on which I installed Condor V 7.8.8.
In the pool there are roughly 50 computers.
For the easiness of the explanation let's assume that there are 100 cores.
I hit the next problem:
2 users want to run simulations in the same time.
User A : effective priority 8.0
User B: effective priority 2.0
User A starts the simulation and occupies 97 cores.
User A has 400 jobs, each of them taking about 5 minutes.
After 5 minutes User B starts the simulation.
User B has also 400 jobs each of them taking about 5 minutes.
User B gets 3 cores.
After 15 minutes User B has still 3 cores.
User B do not get more than 3 cores until User A finish its simulations.
1) Can I do something that User B, which has better priority, to take more cores immediately after one job for User A is finished. With other words, if a core used by User A is released, be be taken by User B, even if User A still has jobs to run?
2) I tried to reset the priorities:
condor_userprio -resetall
Nothing happened. I could only modify the priority factor . Even if the displayed message was
"The accumulated usage was reset for all users"
3) I tried also
condor_userprio -setprio userA 0.5
The message I get is
The priority of userA was set to 0.500000
but when I check the results with condor_userprio I still get the old priority.
All the best,
Daniel