[Swift-user] Teraport question

[Michael Wilde]

I may have asked too soon, as I just found this in the user guide:

"coasterWorkerMaxwalltime specifies the maxwalltime to be used when 
submitting coaster workers. This profile entry is used by the coaster 
execution provider. If this entry is not specified, the coaster provider 
will compute a maxwalltime based on the maxwalltime of jobs submitted. 
(since Swift 0.9) "

Which seems to leave a lot of room for interpretation, but is the 
following reasonable?

If you can, just specify coasterWorkerMaxwalltime, and that will go to 
the resource manager scheduler in the absence of any other times.

But, if you are running jobs on many sites, some of which specify 
coasters and some of which dont, then I take the above to mean that 
coasters will compute a max walltime base on whatever jobs it sees for a 
given site at the moment???? And how does it treat maxWalltime for an 
app (from tc.data) vs. a site?

Can you provide some simple guidelines to make this whole issue of 
queues and times easy to understand and specify?

On 3/19/09 4:34 PM, Michael Wilde wrote:
> The TeraPort queue policies are described at:
> http://www.ci.uchicago.edu/wiki/bin/view/Teraport/QueuePolicies
> 
> which says the fast queue gives you up to an hour.
> 
> -- 
> 
> Separate from that, further clarification of queue and time specs would 
> be helpful.
> 
> There was much discussion on the devel list about how times are treated 
> in coaster scheduling. This left me confused as to whether time specs 
> for coaster jobs were working as desired or still in flux. Specifically:
> 
> 1 - it is unclear if profile time specs are all in hh:mm:ss. I think I 
> reported to swift-devel a coaster case where they were not interpreted 
> in that standard manner.
> 
> 2 - it was unclear whether the time is for the pbs (i.e., 
> coaster-worker) job or for the swift apps that run on them.  Ie how does 
> "coasterWorkerMaxwalltime" which is mentioned in the user guide interact 
> with maxwalltime in the globus profile?
> 
> 3 - is maxwalltime treated the same or different if specified in tc.data 
> vs sites.xml, with respect to coasters?
> 
> 4 - if I specify only a queue but no time limit, do I get the max for 
> that queue? (would hope so...)
> 
> 5 - is job time handling now working as the developers currently intend 
> or are there outstanding issues in this area?
> 
> 
> On 3/19/09 4:07 PM, Ben Clifford wrote:
>> You can run the qstat command to see if your jobs are being queued, 
>> what state they are in, and which queue they went to. For example:
>>
>> 878629.tp-mgt             null             yuechen                0 Q 
>> extended      
>> shows a job submitted by you, that is in state Q (meaning queued) and 
>> that it is in the extended queue.
>>
>> There are a lot of other jobs running on teraport at the moment. You 
>> might have more luck using a different queue, such as 'fast' which you 
>> can specify by adding a line something like this:
>>
>> <profile namespace="globus" key="queue">fast</profile>
>>
>> to your site definition.
>>
>> However, you'll likely still have to wait some time for your jobs to 
>> run - that is the nature of using a job queue...
>>
> _______________________________________________
> Swift-user mailing list
> Swift-user at ci.uchicago.edu
> http://mail.ci.uchicago.edu/mailman/listinfo/swift-user