Thanks Edward. To be honest, most of it comes from compressing the messages. Randbench in this case sends a 16 byte array, with the length of the array it is 20 bytes. So I bet that compression is compressing it to 1-2 bytes for each sync and peer, in this case the benchmark is more ridiculous than an improvement. But it shows that we have too much overhead in sync and the message exchange.
Do you have a less synthetic workload? 2012/7/27 Edward J. Yoon <[email protected]> > I'm testing Hama 0.5 now, it seems almost 3x faster than previous version. > :-) > > http://wiki.apache.org/hama/Benchmarks#Compare_with_old_version > > -- > Best Regards, Edward J. Yoon > @eddieyoon >
