[Swift-devel] Re: devel prios for Mihael

[Michael Wilde]

Consolidating what we discussed on the list into an update on these issues:

> - fix Cobalt provider for general Swift use on Eureka
> (Justin: please file info on this as a bugzilla)

pending word from ALCF.

> - fix general Swift job scheduling to OSG
> 
> -- fix swift-plot-log to give useful reports on runs
> -- target is to support runs for Marc, Glen, and Aashish

pending re-tests by these users on 0.92 w/ changed sites params

> - assist in test and fix of SGE and PBS issues
> 
> -- issues to be gathered and filed

- pending documentation of the issues

> - support Allan to achieve ExTENCI data transfer goals

This is two issues:

HIGH: 1. coasters fails at ~ >60 workers per service

^^^ bubbles up to higher prio

HIGH: 2. throughput w/ provider staging is about 2jobs/sec/service with 3MB data input file per job.  Allan needs at least 10 jobs/sec/service for ExTENCI runs.

Its possible that issue #2 is the same as the next issue below, "coaster time gap issue".

^^^ Justin and Allan are both working to isolate this at the moment.

> - assess coaster time gap issue with Justin; fix if needed

- pending measurement and reproduction of the issue

> - assess coaster polling issue with Justin; fix if needed

- dropped

> - work on runs and plots for Coaster paper, maybe for Swift paper

- on hold pending discussion

> - add option to set coaster-service to passive mode

- low prio. Is its trivial to add?


> - anything else needed for 0.92?

Other support/debug work:

MED Fix problems blocking Jon from reliable large runs.

> 
> Please pipe in with whats missing.
> 
> Thanks,
> 
> Mike
> 
> --
> Michael Wilde
> Computation Institute, University of Chicago
> Mathematics and Computer Science Division
> Argonne National Laboratory

-- 
Michael Wilde
Computation Institute, University of Chicago
Mathematics and Computer Science Division
Argonne National Laboratory