Home
Name Modified Size InfoDownloads / Week
0.1 2013-09-05
docs 2013-09-05
README 2013-09-05 1.1 kB
Totals: 3 Items   1.1 kB 0
Thank you for checking out JUMMP, the Job Uninterrupted Maneuverable MapReduce Platform.

version 0.1
-----------

- Support for the PBS scheduler

- Customizable PBS script to build dynamic maneuverable Hadoop clusters
    * Hadoop Home Directory
    * Internal IP address / Hostname of PBS cluster head node
    * Path to local hard disk for temporary data store
    * Path to directory housing this script (jummp.pbs)
    
- This version requires a dedicated NameNode / TaskTracker outside of scheduler purview

- Usage: qsub -v tj=<JumpTime>,n=<SizeOfCluster>,role='m' jummp.pbs

  * JumpTime: the inter jump time between nodes in minutes
  * SizeOfCluster: number of jumping nodes (DataNode/TaskTracker)
  
- To stop: run kill-jummp.pbs. This will rename of jummp PBS jobs to jummpKill and then 
  qdel them. jummpKill ensures they do not jump to new node 

Citations
---------

Please site our work from Cluster 2013:

W.C. Moody; L. Ngo; E. Duffy; A. Apon; "JUMMP: Job Uninterrupted Maneuverable 
MapReduce Platform", Cluster Computing (CLUSTER), 2013 IEEE International 
Conference on , 23-27 Sept. 2013
Source: README, updated 2013-09-05