Subversion Repositories cctools

[/] [branches/] [lyu2/] - Rev 1996

Rev

Go to most recent revision

Filtering Options

Rev Log message Author Age Path
1751 merge from trunk r1743:1750 lyu2 459d 12h /branches/lyu2/
1750 added feature: By default a worker executes a task in its child process. Both the new master and new the worker are backward compatible.

ongoing feature: worker selecting a master based on a given probability.
lyu2 459d 12h /branches/lyu2/
1746 Refactored the work queue accounting code.
Makeflow can turn on the WQ master capacity related functionalities by options.

On-going features: worker executing a task in a separate process.
randomized master selection according to a given worker distribution.
lyu2 460d 09h /branches/lyu2/
1744 merge from trunk r1674:1743 lyu2 462d 12h /branches/lyu2/
1675 merge from trunk -r 1647:1674 lyu2 485d 12h /branches/lyu2/
1666 merge from trunk r1435:1647 lyu2 486d 12h /branches/lyu2/
1647 batch_job_local.c: uses exec, instead of system(), to run a job as a separate local process.
work_queue_pool: fixed "not removing all workers" on ctrl-c
lyu2 506d 06h /branches/lyu2/
1462 1. work_queue_pool: added automatic worker pool provisioning for matching masters.
2. reorganized code for merging back to the trunk (more clean up needed though)
lyu2 604d 16h /branches/lyu2/
1438 merge from trunk r1125:1435

cctools compiles successfully.
Makeflow is able to pass all of its own unit tests.
As I've tested, makeflow with 'work queue' option, work_queue_pool, work_queue_worker, and work_queue_status work correctly. And these covers most of the code that I've touched in my wq-accounting branch.
lyu2 625d 11h /branches/lyu2/
1435 rename worker.c to work_queue_worker.c again, for merging the trunk to my branch lyu2 626d 14h /branches/lyu2/
1431 undoing changes committed in r1430 lyu2 627d 05h /branches/lyu2/
1430 renamed worker.c in my branch lyu2 627d 06h /branches/lyu2/
1429 code that worked for the cloudcom11 paper. lyu2 627d 11h /branches/lyu2/
1406 Added a script for running master location experiments (for worker pool management paper, cloudcom11) . lyu2 647d 11h /branches/lyu2/
1405 A decent worker removal algorithm for work_queue_wait_adaptive.

1.For master capacity calculation, the average application time is computed over all finished tasks (with outliers removal, i.e., numbers that are 10 times larger than avg is discarded, this renders more stable results for small workflows).

2.Incorporated master capacity stand deviation into the worker removal algorithm.

3.Added capacity tolerance option to makeflow.
lyu2 647d 11h /branches/lyu2/
1394 added more curves in graphs generated by work_queue_benchmark.sh lyu2 659d 18h /branches/lyu2/
1393 added "work queue wait routine -- adaptive" option lyu2 659d 18h /branches/lyu2/
1392 added excessive worker removal functionality. This version is supposed to work with work_queue_wait_adaptive routine. It works, but I'm still tweaking the worker removal algorithm in order to make it work better. lyu2 659d 18h /branches/lyu2/
1339 refactored master capacity measurement code.
added a new work_queue_wait routine: work_queue_wait_adaptive, which is supposed to make a smarter decision on whether to start a new task on a brand new worker or an old worker, without much computational overhead. Still experimenting on this routine. So, don't rely on its code in this version.
lyu2 673d 14h /branches/lyu2/
1290 separated the running experiments and generating graphs procedures.
added number of workers provided in each generated graph.
lyu2 704d 14h /branches/lyu2/

1 2 Next >

Show All