Temporarily, I'll fix this by adding simple logic which is computes the number of global synchronization steps based on the maximum split size and heap of bsp child.
P.S., My summer vacation: 11 ~ 19, Aug. On Wed, Jul 11, 2012 at 11:12 AM, Edward J. Yoon <[email protected]> wrote: > This quick-patch was for avoiding huge memory usage (commodity PCs > cluster) but, seems I made a bug. > > Since each peer loads different split data, some tasks start bsp() > earlier than others. > > Will fix on HAMA-599 > > On Wed, Jun 27, 2012 at 5:19 PM, <[email protected]> wrote: >> Author: edwardyoon >> Date: Wed Jun 27 08:19:17 2012 >> New Revision: 1354365 >> >> URL: http://svn.apache.org/viewvc?rev=1354365&view=rev >> Log: >> Increase sync interval >> >> Modified: >> hama/trunk/graph/src/main/java/org/apache/hama/graph/GraphJobRunner.java >> >> Modified: >> hama/trunk/graph/src/main/java/org/apache/hama/graph/GraphJobRunner.java >> URL: >> http://svn.apache.org/viewvc/hama/trunk/graph/src/main/java/org/apache/hama/graph/GraphJobRunner.java?rev=1354365&r1=1354364&r2=1354365&view=diff >> ============================================================================== >> --- hama/trunk/graph/src/main/java/org/apache/hama/graph/GraphJobRunner.java >> (original) >> +++ hama/trunk/graph/src/main/java/org/apache/hama/graph/GraphJobRunner.java >> Wed Jun 27 08:19:17 2012 >> @@ -429,7 +429,7 @@ public final class GraphJobRunner<V exte >> vertex.runner = this; >> >> lines++; >> - if((lines % 50000) == 0) { >> + if((lines % 100000) == 0) { >> peer.sync(); >> GraphJobMessage msg = null; >> while ((msg = peer.getCurrentMessage()) != null) { >> >> > > > > -- > Best Regards, Edward J. Yoon > @eddieyoon -- Best Regards, Edward J. Yoon @eddieyoon
