tag:blogger.com,1999:blog-7181711759016870742.post1078282269175368691..comments2022-05-10T12:26:08.070+02:00Comments on Thomas Jungblut's Blog: Graph Exploration with Apache HamaThomas Jungbluthttp://www.blogger.com/profile/07157841886768146088noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-7181711759016870742.post-63394192543748396772012-03-13T07:27:12.888+01:002012-03-13T07:27:12.888+01:00A modified version of this was 10 times faster for...A modified version of this was 10 times faster for 100mio vertices than MapReduce for around 50 iterations. <br />A state of the art implementation will get rid of the second barrier sync and distributes counters through Zookeeper. <br />Once https://issues.apache.org/jira/browse/HAMA-515 hits trunk this will be much more easier to implement.<br /><br />This example is a bit screwed because one has to partition the dataset and do the whole IO. This is now covered by Hama itself.<br /><br />This problem can be much more easier solved by BSP than MapReduce, at least it is much more intuitive. The speed improvement is just a nice side-effect making Hama useful as a real-time system.Thomas Jungbluthttps://www.blogger.com/profile/07157841886768146088noreply@blogger.comtag:blogger.com,1999:blog-7181711759016870742.post-53606682297398572542012-03-12T22:58:55.165+01:002012-03-12T22:58:55.165+01:00So how did this BSP implementation perform compare...So how did this BSP implementation perform compared to the MapReduce version? Faster? <br />Why would somebody use BSP implementation, if the problem could be solved easily by MapReduce?Liyanghttps://www.blogger.com/profile/12236724639393897960noreply@blogger.com