[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Condor-users] Flocking...

> Hello all,
> After I got the answer that Condor doesn't know to work with 
> more then 1 
> network interface , I setup two masters on my master machine, 
> and configure 
> each to have its own pool.
> So, I have public-pool , and private-pool. the public one can 
> communicate 
> with the public stations that submit the job, and the 
> private-pool is the 
> one that do all the work.
> I also configured the public-pool that will flock_to private-pool ( 
> flock-from ).
> I can submit jobs to public-pool, but this one doesn't "forward" the 
> request to the private-pool.
> Do you know a way I can get information on this flocking mechanism ?
> How can I check if the public-pool really tries to flock to 
> the private-pool ?

You need it setup so that each submit node can "talk" to each execute node.
Flocking forwards the requests to the other Pool, but the final comms still
has to take place between submitter and executer and vice versa.

Since I presume your public pool nodes cannot address nodes directly in the 
private pool, then this won't work.

Possible solutions are:
* Putting everything in one pool and making a machine on the "border" (presumably
the current central node of your private pool) a single submit node.
* Putting your private pool on a public network (possibly in addition
to its private one)
* Wait a while, the Condor team have been working on a technique called
a GCB which will allow these communications to be delegated to a machine accessible
to all.
* Setup your network so that any requests to the "hidden" nodes go through the 
master node and vice versa (I am not a network guy, but this may be possible,
albeit not so efficient).