Hi Walker, The old contract was changed because if a task gets re-allocated on a different machine on failure, that can change the peer-Ids of all the tasks. The current trunk version returns the task-id, which remains same throughout the lifecycle of the job execution. With this peerIndex could be made a part of your computation logic.
-Suraj On Tue, Sep 18, 2012 at 11:59 AM, Thomas Jungblut <[email protected] > wrote: > Hi Walker, > > seems the contract is broken and the array is not sorted anymore. > > Try to use: > > String[] names = peer.getAllPeerNames(); > Arrays.sort(names); > int index = Arrays.binarySearch(names, peer.getPeerName()); > > Sorry for the inconvenience, we have replaced it with the task id in > the latest trunk, so this will be fixed in the next release. > > 2012/9/18 顾荣 <[email protected]>: > > Hi, > > > > I have a Hama cluster of 3 nodes. The host name of the nodes are > slave019, > > slave020, slave021. I wirte a simple program to test the PeerIndex and > > PeerName info in Hama. > > > > In setup function, I choose a peer as master, and the in bsp function > each > > peer get its PeerIndex and its PeerName by peer.getPeerIndex() and > > peer.getPeerName() API and send it to the master peer. Lastly, in > cleanup > > function of the master peer, it write out the messages in its queue. The > > result is below: > > > > peerName: slave020:61001 peerIndex: -7 > > peerName: slave019:61002 peerIndex: 4 > > peerName: slave021:61001 peerIndex: 6 > > peerName: slave019:61001 peerIndex: -1 > > peerName: slave021:61002 peerIndex: -8 > > peerName: slave020:61002 peerIndex: -7 > > peerName: slave019:61003 peerIndex: 5 > > peerName: slave021:61003 peerIndex: 7 > > peerName: slave020:61003 peerIndex: -7 > > > > The peerNames are reasonable, but I am confused with the peerIndex > > information. What does they mean? According to the source code of > > BSPPeer.java, the comment message is below > > /** > > * @return the index of this peer from sorted array by name. > > */ > > public int getPeerIndex(); > > > > However, the peerIndex messages I got seem not to be so....Can you give > me > > a hint? > > > > Thanks in advance. > > > > Walker >
