Commit graph

228 commits

Author SHA1 Message Date
Pradyumna Kaushik
95c419785f Made sure that capping is done only if a new cap value is computed. Otherwise we don't unnecessarily cap the cluster. This reduces overhead. 2017-01-08 19:08:46 -08:00
Pradyumna Kaushik
59e08f0a1a fixed spacing 2017-01-08 19:08:02 -08:00
Pradyumna Kaushik
15d3aa2177 removed commented code 2017-01-08 19:01:08 -08:00
Pradyumna Kaushik
f46cd32869 Bug fix to start capping when the first task fits an offer. 2017-01-08 18:59:51 -08:00
Pradyumna Kaushik
2744bbe0f4 Made sure that the capping happens only if a different cap value has been determined. Otherwise we don't unncessarily cap again. This reduces overhead. 2017-01-08 18:58:00 -08:00
Pradyumna Kaushik
d9f626eefb BinPackedMaxMin with piston capping 2017-01-06 23:40:30 -08:00
Pradyumna Kaushik
757684d3a4 BinPackedMaxMin scheduler with Proactive cluster wide capping 2017-01-06 23:40:05 -08:00
Pradyumna Kaushik
154bacdf7f Added missing line -- Add task to list of running tasks 2017-01-06 21:05:56 -08:00
Pradyumna Kaushik
a0f911e4e4 Used composition to use common functions from base.go. Also incorporated schedTrace and made the necessary changes to the constructor 2017-01-06 16:49:00 -08:00
Pradyumna Kaushik
722bd233e3 Used composition to use common functions from base.go. Also incorporated schedTrace and made the necessary changes to the constructor 2017-01-06 16:47:59 -08:00
Pradyumna Kaushik
20efe92525 Piston Capping with BinPackedSortedWattsClassToWatts 2017-01-06 16:01:45 -08:00
Pradyumna Kaushik
678c113f0a Proactive Clusterwide Capping with BinPackedSortedWattsClassToWatts 2017-01-06 16:01:18 -08:00
Pradyumna Kaushik
d5860a91ae fixed merge conflicts 2017-01-06 15:58:50 -08:00
Renan DelValle
841c5e0392 Minor comments and style changes 2017-01-06 18:53:03 -05:00
Pradyumna Kaushik
270b5a7ee1 Fixed bug. Now watts is taken from task.ClassMapWatts[nodeClass] and not task.Watts. Changed the function signature of newTask(...) to take in another parameter called newTaskClass that corresponds to the nodeClass. 2017-01-06 15:41:13 -08:00
Pradyumna Kaushik
0135ee69e4 Added TODO to make the newTask(...) variadic 2017-01-06 15:39:12 -08:00
Renan DelValle
776bffb100 Finished bin packed max min. 2017-01-04 20:25:54 -05:00
Renan DelValle
2678032c2c Go fmt on the entire project 2017-01-03 20:57:25 -05:00
Renan DelValle
b636ff490c Using anonymous and type promotion to code share more efficiently 2017-01-03 20:56:55 -05:00
Renan DelValle
3f04e4cc4c Minor fixes. Updated the README. 2017-01-03 18:10:05 -05:00
Renan DelValle
312728a2f6 Adding node class mapping for Watts as a resource in binpack 2016-12-23 21:04:15 -05:00
Pradyumna Kaushik
b25158336d Used the capping strategy 'proactiveclusterwidecappers' from pcp/ instead of from the same package as it was moved. 2016-12-22 23:20:24 -05:00
Pradyumna Kaushik
7d0a902773 Used the capping strategy 'proactiveclusterwidecappers' from pcp/ instead of from the same package as it was moved 2016-12-22 23:20:24 -05:00
Pradyumna Kaushik
f829cfea66 Moved it to pcp/ as it is not a scheduler and is a capping strategy 2016-12-22 23:20:18 -05:00
Renan DelValle
71c889982f resolved merge conflicts 2016-12-22 23:17:01 -05:00
Pradyumna Kaushik
2d3b484151 Using ranked proactive cluster wide capper with clever recapping 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
f55169c91c Fixed corner case bug with sorting of tasks. Not sorting if there are no more tasks to schedule. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
7bd562cacb synchronized operations on tasksRunning and hence, prevented previously occuring race condition. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
24d2b89aa0 synchronized operations on tasksRunning and hence prevented the previously occuring race condition. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
89a294b365 Removed TODO comment in ResourceOffers(...), that wasn't necessary anymore. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
fae3a35403 Added to TODO (generic running average computation). 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
50d1d79051 fixed naming convensions to be camel cased. Reformatted the code. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
55ea017a9a Removed a commented line. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
6b32c9eeda Added the proactive cluster wide capping with ranked tasks scheduler to the list of schedulers. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
72bd54374c Added proactive cluster wide capping with ranked tasks as another scheduler. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
81d795f197 formatted the code 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
f6ec974182 changed the keys in taskMonitor from offer.SlaveId() to offer.Hostname. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
a255351952 fixed bug in cleverRecap(...). Now we switch from the primitive recap to the clever recap as the cap determined by the later would be lesser when the cluster is relatively idle. 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
8bbecf9d5f Markdown fix in README 2016-12-22 22:59:03 -05:00
Pradyumna Kaushik
5329fce402 Fixed an error in markdown 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
e5aaf2dbe9 README to list the different scheduling algorithms defined 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
87e2c802dd Praoctive cluster wide capping after ranking the tasks based on the requested watts 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
4bc81707e0 Added another line that needs to be uncommented to choose cleverRecap. 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
ec4f4e0f03 removed rankedDetermineCap(...) as it was not needed. This algorithm has been integrated into proactiveclusterwidecappingranked.go 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
c1eaa453a2 Sycnrhonized operations that change the value of the cluster wide cap. Added cleverRecap(...) that determines the recap value of the cluster at a much finer level, taking into account the average load on each node in the cluster. Bug fix in cap.go -- closed the session once capping had been done. This prevented from running out of file descriptors. 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
b7394b8762 formatted the code 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
4d13c432c4 Kept track of totalPower per node. The watts resource for the first offer corresponds to the total power per node. Removed tasks, that had all their instances scheduled, from the list of tasks to schedule. Also, calling recap(...) every time a task completes to determine the new cluster wide cap." 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
3551de20da The computation of the cluster wide cap now considers total power per node rather than the available power per node. Also, added function recap(...) that is called to compute the cluster wide cap once a task completes. This value is used to change the cluster wide cap once a task completes." 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
ed25adcc6c Made a check to see whether cluster wide capping has started and if not then starting the go routine that performs the cluster wide capping at regular intervals. 2016-12-22 22:59:02 -05:00
Pradyumna Kaushik
7b19cfb4fc fixed an error 2016-12-22 22:59:02 -05:00