ec2 backend sends multiple jobs to same VM

[GoogleCodeExporter]

Reported via email by Joel Fiddes:

  I often seem to get multiple jobs (i.e., simulation boxes) being sent
  to the same VM. This can cause memory problems. Sometimes jobs from
  two separate calls maybe assigned to the same VM.

  Consider 2 calls:

  SESSION1 (S1): gtsub_control -C 10 -o ~/sim/gst200_811 -s ~/sim/gst200_811 ~/sim/gst200_811 -m 14GB 8:11
  SESSION2 (S2): gtsub_control -C 10 -o ~/sim/sd200_811 -s ~/sim/sd200_811 ~/sim/sd200_811 -m 14GB 8:11

  For example, situations I've seen:
  (a) B8 S1 and B9 S2 (or more) assigned to same VM
  (b) B8:10 S1 assigned to same VM

  the VM flavor is m1.xlarge (14GB, 8cores)

Original issue reported on code.google.com by riccardo.murri@gmail.com on 18 Jul 2013 at 10:02

[GoogleCodeExporter]

On 18 July 2013 11:21,  <joel.fiddes@gmail.com> wrote:
| I often seem to get multiple jobs (i.e.
| simulation boxes) being sent to the same VM.

This is actually normal in GC3Pie: it tries to pack jobs into existing
VMs, as long as the total job requirements are compatible with the VM
flavor.

In other words, if you are using a m1.xlarge VM, with 14GB of RAM, and
your jobs request 7GB of RAM each, GC3Pie will try to start *two* jobs
on the same VM (even if they belong to different sessions).

This is only a problem if:

1- GC3Pie start more than two jobs on the same VM;
2- the jobs actually use more RAM than requested.

Which of the two cases is happening here?

Original comment by riccardo.murri@gmail.com on 18 Jul 2013 at 11:35

[GoogleCodeExporter]

I could not reproduce this issue, using your configuration file and
the `gtsub_control` command-line that you sent me.

My guess at what happens is this:

- the memory requirement for a job is set when the job is *first
  created*;

- jobs are created when the session is created;

- the default memory setting in `gtsub_control` is 7GB;

- so if you ever call `gtsub_control ...` and forget to add `-m 14GB`,
  then jobs will be created with a 7GB requirement, no matter if you
  later stop and restart with a `-m 14GB` appended.

- with a 7GB requirement, two jobs would fit into one m1.xlarge machine.

So, can you try submitting new sessions, paying attention that `-m
14GB` is there from the start?

If you still have a couple of old sessions that exhibited the behavior
you complained about, would you please run this command:

    ginfo -vv -s "session_name" >> /tmp/"session_name".txt

and then send me the files /tmp/session_name.txt (replace "session_name"
with the actual session name).

Original comment by riccardo.murri@gmail.com on 19 Jul 2013 at 3:59

[GoogleCodeExporter]

Not having received the requested information and not being able to reproduce 
the issue, I will close this issue.

Please feel free to re-open it the problem is found again.

Original comment by antonio....@gmail.com on 28 Nov 2013 at 2:40

• Changed state: Invalid