| T O P I C R E V I E W |
| deshmukh |
Posted - Nov 15 2007 : 11:17:23 AM
The cluster now supports interactive jobs that can be used for designing fsf files. Instructions on these jobs is here http://fourier.biac.duke.edu/wiki/doku.php/biac:cluster:interactive .
Also, the cluster now implements a "fair use" policy that balances user resource allocation over time. This implies that you no longer have to self-impose a job submission policy eg: waiting to submit a large number of jobs or submitting only few at a time. Jobs waiting in the queue will be automatically sorted to uniformly distribute job priorities between user's based on historical resource consumption.
|
| 15 L A T E S T R E P L I E S (Newest First) |
| josh.bizzell |
Posted - May 28 2008 : 3:22:52 PM
Lihong,
Sometimes when all nodes are maxed out by the users queue, the interact queue will not allow people to log in. I'm not exactly sure why this happens, but as you said, if you keep trying, it will eventually let you log in.
If you have a new experiment and you cannot access it, let me know. Most likely, I'll just need to restart the mount daemon on the machines (which I will do right now).
-Josh |
| wang |
Posted - May 28 2008 : 3:12:10 PM
I received the same message as John had, but I was able to login qinteract after a while. So I guess there does seem to have some job limitation for qinteract. With the plan of retiring golgi, are we increasing the usability of qinteract from the Cluster?
When I was trying to list my study: lnexp CogDEP.01, I got the ERROR: Unable to mount experiment. How do I get a new experiment registered in qinteract?
Many thanks. |
| clithero |
Posted - Mar 06 2008 : 5:00:28 PM
I just tried to open qinteract on the cluster and got the following response:
Your "qrsh" request could not be scheduled, try again later.
This has happened three times for me. Is there a limit to the number of interactive jobs?
Thanks,
John |
| francis.favorini |
Posted - Jan 16 2008 : 11:24:24 PM
We currently have one node to install. We are waiting for power installation in the machine room.
-Francis |
| dvsmith |
Posted - Jan 16 2008 : 8:59:37 PM
I've heard more nodes have been ordered. When will these be installed?
-David |
| josh.bizzell |
Posted - Jan 16 2008 : 3:16:02 PM
Due to the large number of waiting jobs, I'm going to enable node4 to accept submissions from the users queue. This means that during peak times, all four processors will be in use running submitted jobs. You should still be able to use the interact queue, but it might be a little slower. If anyone experiences performance issues during interactive jobs, please let me know as soon as possible.
Thanks,
Josh
|
| josh.bizzell |
Posted - Jan 10 2008 : 2:40:06 PM
As mentioned on the wiki page, sometimes it takes a couple of minutes (2-5 min) for the interact queue to exit.
If, in fact, it doesn't exit after this amount of time, then check that there aren't any open applications accessing node4 (as David mentions above).
If you see that you have a "pending" interactive job that you're no longer using, you can use "qdel" to remove this job from the queue list.
-Josh |
| dvsmith |
Posted - Jan 10 2008 : 2:34:44 PM
That might work, but first make sure you don't have any active windows (e.g., the FEAT GUI or a gedit window) that were running from the qinteract job. If you close them and if you've typed 'exit' already, you should get logged off unless something else is happening.
-David |
| tankersley |
Posted - Jan 10 2008 : 2:05:03 PM
If "exit" is not logging us out of qinteract on node4, can we use a separate terminal window to try to
qdel jobnum for the qinteract session #?
Thanks,
Dharol |
| tankersley |
Posted - Jan 03 2008 : 3:12:45 PM
When i type
qinteract
I am prompted for my password, which is denied. I've tried both the password i set for the cluster, as well as the original password i was sent for the cluster. Is there an additional step i need for access?
Thanks,
Dharol |
| mullette-gillman |
Posted - Dec 10 2007 : 10:20:17 AM
Thank you! |
| josh.bizzell |
Posted - Dec 07 2007 : 4:52:49 PM
As Chris mentioned, interactive jobs are "run" on node4 of the cluster, not the head node. However, users shouldn't run jobs when logged into node 4 via qinteract (some FSL functionality has been disabled anyway).
We have also removed node4 from the users queue, so you'll see that node 4 (as well as nodes 1 & 2, for different reasons) never accepts jobs from qsub.
-Josh |
| petty |
Posted - Dec 07 2007 : 4:37:15 PM
they do infact use processing power, but since you can't actually run a job through qinteract it shouldn't really be noticable. But just because there are 4 processors, doesn't mean that only 4 jobs can run on node 4 through qinteract.
and yes, the que system does have prior knowledge, but i don't think it should affect the way things are running currently. if we had hundreds of users submitting 100s of jobs per day is when you would notice. the cluster thus far isn't all that busy. |
| mullette-gillman |
Posted - Dec 07 2007 : 4:26:58 PM
Does anyone know for certain whether interactive jobs take up a processor on the cluster? Do all open interactive jobs run off the head node?
Also, Does the que-ing system really take into account prior use? I have yet to see any evidence of jobs being entered into processors out of the order that they are initially submitted (I've been watching for it)?
Thanks!
O'Dhaniel |
| wang |
Posted - Nov 20 2007 : 5:17:17 PM
Thanks Imran. Now I got it successfully. Indeed I have tried to mount twice. I didn't realize that the ATD.02 is under the experiment direcotry. Thanks again. |
