tag:blogger.com,1999:blog-7994087232040033267.post471839538167791257..comments2024-03-03T00:23:26.457-08:00Comments on Pragmatic Programming Techniques: Google Pregel Graph ProcessingRicky Hohttp://www.blogger.com/profile/03793674536997651667noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-7994087232040033267.post-72863966856175508622013-04-14T02:48:10.026-07:002013-04-14T02:48:10.026-07:00who handles synchronization part and how?who handles synchronization part and how?!-!aR$#!Lhttps://www.blogger.com/profile/00747802106564599923noreply@blogger.comtag:blogger.com,1999:blog-7994087232040033267.post-89292658410125973362010-07-19T23:27:40.834-07:002010-07-19T23:27:40.834-07:00We can have multiple masters and use something lik...We can have multiple masters and use something like a Zookeeper to elect a leader master.<br /><br />Yes, worker are stateless while the partition are store in a DB server that other workers can access.<br /><br />Rgds,<br />rickyRicky Hohttps://www.blogger.com/profile/03793674536997651667noreply@blogger.comtag:blogger.com,1999:blog-7994087232040033267.post-36320946496947609732010-07-16T16:06:00.221-07:002010-07-16T16:06:00.221-07:00Thank you for the great post.
Is "master"...Thank you for the great post.<br />Is "master" a single point of failure? I think that in any case not much harm can be done if it is down, just the workers will not be able to proceed to the next superstep. <br />But what are protocols to select a new master? Also it looks like one implicit assumption is that the storage is shared and other workers can access stored state (checkpoint) of the failed worker. Is it right?Walter Bogoradhttps://www.blogger.com/profile/08725349627112465457noreply@blogger.com