Unfortunately there are still some problems with communications.
I didn't get any error likes connection loss exception, but I'm sending
message with tag:
byte[] tagName = Bytes.toBytes("TEST_TAG");
and once (only!) during my experiment I received something like:
String msgTag = Bytes.toString(received.getTag());
// msgTag = "[B@56c163f"It looks like sometimes messages are corrupted. Cheers, Pawel PS. It could be great to see your benchmark results. 2011/2/16 Edward J. Yoon <[email protected]> > I decided to add a "random communication benchmark" tool. In this week > (or next week), I'll share with you my benchmarking experience. I have > 20 (160 cores) servers. > > Thanks. > > 2011/2/16 Edward J. Yoon <[email protected]>: > > Looks like problem of sync. Can you try again it after add > Thread.sleep(100); line? > > > > Sent from my iPhone > > > > On 2011. 2. 16., at 오후 3:24, Paweł Brach <[email protected]> wrote: > > > >> Yes, I have of course. My cluster has been configured and both examples > >> PiEstimator and SerializePrinting work (there is communication between 3 > >> nodes). I've modified your example - PiEstimator (put everything in the > >> loop) and it works for few iterations (there is communication) and after > >> that connection is lost. After that connection is re-established but > some > >> messages are missing. It looks like that Hama framework is very unstable > >> when it's loaded and many messages are sending between nodes. > >> On the same cluster I've configured Apache Hadoop and it's very stable. > >> If you have own cluster configured, could you run my example on it ? > Have > >> you ever run something more complicated than PiEstimator and > >> SerializePrinting on it ? > >> > >> Cheers, > >> Pawel > >> > >> 2011/2/16 Chia-Hung Lin <[email protected]> > >> > >>> Have you configured zookeeper in hama-site.xml? Hama makes use of > >>> zookeeper to do node communication IIRC. > >>> > >>> Opening socket connection to server cl5/127.0.1.1:2181 > >>> > >>> indicates that seems only localhost is up. If this is the case, you > >>> can change hama.zookeeper.quorum property pointing with value set to > >>> e.g. > >>> > >>> <property> > >>> <name>hama.zookeeper.quorum</name> > >>> <value>node1,node2,node3,node4,node5</value> > >>> </property> > >>> > >>> Hope it helps > >>> > >>> 2011/2/15 Paweł Brach <[email protected]>: > >>>> Hello, > >>>> > >>>> During last few days I've tested Hama solutions and today I found some > >>>> strange error in Hama framework. If you run a simple job with more > than > >>> few > >>>> supersteps the following error occures: > >>>> > >>>> 2011-02-15 15:13:55,934 ERROR org.apache.hama.bsp.BSPPeer: > >>>> 2011-02-15 15:13:56,525 INFO org.apache.zookeeper.ClientCnxn: Opening > >>> socket > >>>> connection to server cl5/127.0.1.1:2181 > >>>> 2011-02-15 15:13:56,526 WARN org.apache.zookeeper.ClientCnxn: Session > 0x0 > >>>> for server null, unexpected error, closing socket connection and > >>> attempting > >>>> reconnect > >>>> java.net.ConnectException: Connection refused > >>>> at sun.nio.ch.SocketChannelImpl.checkConnect(Native Method) > >>>> at > >>>> sun.nio.ch.SocketChannelImpl.finishConnect(SocketChannelImpl.java:592) > >>>> at > >>>> org.apache.zookeeper.ClientCnxn$SendThread.run(ClientCnxn.java:1078) > >>>> 2011-02-15 15:13:56,626 ERROR org.apache.hama.bsp.BSPPeer: > >>>> org.apache.zookeeper.KeeperException$ConnectionLossException: > >>>> KeeperErrorCode = ConnectionLoss for /bsp > >>>> > >>>> You can reproduce that by running PiEstimator (the newest source code > >>> from > >>>> svn) with small changes - put whole body of the bsp() method in the > for > >>>> loop. So add in the beginning following line: > >>>> > >>>> for (int j = 0; j < 100; j++) { > >>>> // oryginal bsp() code > >>>> } > >>>> > >>>> When I'm trying to run it, the framowork hangs and mentioned before > error > >>>> occures. > >>>> > >>>> Your help will be appreciated. > >>>> > >>>> Cheers, > >>>> > >>>> -- > >>>> Pawel Brach > >>>> > >>> > >>> > >>> > >>> -- > >>> ChiaHung Lin @ nuk, tw. > >>> > >> > >> > >> > >> -- > >> Paweł Brach > > > > > > -- > Best Regards, Edward J. Yoon > http://blog.udanax.org > http://twitter.com/eddieyoon > -- Paweł Brach
