AFAIR raid10 should be around 2x faster on both read and write operations comparing to a simple jbod.
Sent from my iPad On 2010-09-26, at 10:09 PM, Tao Xie <[email protected]> wrote: > Thanks, Ryan. But in the paper they said their Read operations retrieves an > entire record. And my nodes get even more memory than theirs. They has only > 8GB ram while I got 24GB. The only difference is I use JBOD rather than > RAID10 array. I think JBOD would be better for HDFS than RAID. Is that > right? > > > 2010/9/27 Ryan Rawson <[email protected]> > >> hey, >> >> there are a lot of settings which all affect speed, you might want to >> make sure you are running the exact config the paper is. for example >> "readallfields".. if you have a wide column setting it to 'true' will >> go much slower (since you are returning a lot more data) than if you >> set it to false. Also there are known bottlenecks in our HTable >> client, which exist to reduce the # of socket connections, but you can >> get higher performance with multiple VMs, although that is more >> complex. >> >> Also dont forget hardware.. .make sure your HW is the same as the papers. >> >> Good luck! >> -ryan >> >> On Sat, Sep 25, 2010 at 8:19 AM, Tao Xie <[email protected]> wrote: >>> I want to reproduce the results in the ycsb paper. I run hbase 0.20.6 >> and >>> hadoop 0.20.2. My cluster is like this: >>> >>> 1 Node as HMaster + ZK >>> >>> 6 Nodes as DN, RS >>> >>> 1 Node as Hbase client. >>> >>> I think this environment is something like the one used by the paper. >>> >>> When I run tests like workloadb with 100 threads, I get at most 2500 >> ops/sec >>> throughput and read latency is about 40~50 ms, which is much higher than >> the >>> paper results (about 10+ ms). I wonder if anybody is running ycsb too and >>> can give me some hints. >>> >>> >>> >>> Thanks in advance. >>> >>
