Distributed autograding in Docker containers about tango HOT 6 CLOSED

Why does it need to be one job at a time on each machine?

from tango.

I think that we discussed that this would be the "simpler" option. There's no reason why we wouldn't want to eventually make use of all cores.

from tango.

We could make use of all cores by simply "prealloc'ing" one docker instance
for each core.

On Saturday, March 28, 2015, Jake Zimmerman [email protected]
wrote:

I think that we discussed that this would be the "simpler" option. There's
no reason why we wouldn't want to eventually make use of all cores.

—
Reply to this email directly or view it on GitHub
#52 (comment).

Sent from my iPhone.

from tango.

I don't think there's an analog to prealloc in Docker. If you have a job
with Docker, you just run it; there's no need to prealloc.

On Sat, Mar 28, 2015 at 11:50 PM droh [email protected] wrote:

We could make use of all cores by simply "prealloc'ing" one docker instance
for each core.

On Saturday, March 28, 2015, Jake Zimmerman [email protected]
wrote:

I think that we discussed that this would be the "simpler" option.
There's
no reason why we wouldn't want to eventually make use of all cores.

—
Reply to this email directly or view it on GitHub
#52 (comment).

Sent from my iPhone.

—
Reply to this email directly or view it on GitHub
#52 (comment).

from tango.

The Tango scheduler, which is independent of any particular vmms, is designed
around the idea of matching a job with a free instance. So to get the same
behavior as our current Tashi system, we would simply prealloc 40 (5
backend servers x 8 cores per sever) and then round robin schedule.

On Saturday, March 28, 2015, Jake Zimmerman [email protected]
wrote:

I don't think there's an analog to prealloc in Docker. If you have a job
with Docker, you just run it; there's no need to prealloc.

On Sat, Mar 28, 2015 at 11:50 PM droh <[email protected]
javascript:_e(%7B%7D,'cvml','[email protected]');> wrote:

We could make use of all cores by simply "prealloc'ing" one docker
instance
for each core.

On Saturday, March 28, 2015, Jake Zimmerman <[email protected]
javascript:_e(%7B%7D,'cvml','[email protected]');>
wrote:

I think that we discussed that this would be the "simpler" option.
There's
no reason why we wouldn't want to eventually make use of all cores.

—
Reply to this email directly or view it on GitHub
#52 (comment).

Sent from my iPhone.

—
Reply to this email directly or view it on GitHub
#52 (comment).

—
Reply to this email directly or view it on GitHub
#52 (comment).

Sent from my iPhone.

from tango.

The PR for this: #69

from tango.