That makes sense, and I initially thought it was your submission file
that may be big. Such a large output would crash the controller very
easily, if it creates a property list file as it seems it does from
the crash log. The property list will be built in memory and will be
at least 1 GB large. Creating several files like this will quickly
reach the 2 GB limit (probably 1 or 2 files will be enough!).
Given the 2 GB limit, it is probably reasonable to limit yourself to
<100 MB size files. But you also have to consider that the file is
going to first be sent by the agent to the controller, then by the
controller to the client. If client/controller are the same machine,
maybe smart things happen, but the overhead might be still there.
Using a solid separate fileserver is probably a better idea.
charles
On Feb 22, 2007, at 11:24 AM, Adam Kocoloski wrote:
Hi all,
So, the job I was submitting produce O(1GB) of binary output files,
and I'm certain that the controller resource usage and crashes were
due to trying to handle all that data. I tried adding a few lines
to the end of each task where I AFP guest-mount a "drop box" on a
fileserver and move my output files there instead of bringing them
back through the controller. Everything is rock-solid now, and the
xgridcontrollerd memory usage is stable at about 7MB.
I guess this is not surprising. In working with other batch
systems (Condor, LSF, SGE) I've never let the system handle
anything but the streams. It'd be nice if we could quantify just
how much data is reasonable to bring through the controller. Regards,
Adam
On Feb 22, 2007, at 6:50 AM, Adam Kocoloski wrote:
Hi Charles,
On Feb 22, 2007, at 12:30 AM, Charles Parnot wrote:
how big is your batch xml file?
904 KB
Is it crashing at submission or while running the jobs?
While running the jobs. First time it crashed I had a new job
Pending with 0 tasks in Xgrid Admin (plus a couple other running
jobs), but subsequent crashes occurred during running.
--Adam _______________________________________________
Do not post admin requests to the list. They will be ignored.
Xgrid-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Xgrid-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden
