Anyone had a rough measurement of random read/write perf and throughput? Assume a typical machines/workload, the region server has 5GB for memtable, further assume each key (20 bytes) has 100 bytes value (for simplicity, just one cf, one column). Further assume workload is against a single region on this region server. What is the latency/throughput of write-only workload? (assume all are update ops, no insert ops, and updates on different keys are uniformly distributed) What is the latency/throughput of read-only workload? (assume lookup on different keys are uniformly distributed)
Really want to get more sense of perf number of hbase for key/value scenario. Sorry, further assume clients are multi-processed, multi-threaded, which can generate large enough concurrent requests before the network is saturated. best, -zhimao On Wed, Sep 26, 2012 at 9:14 PM, Kevin O'dell <[email protected]>wrote: > -scm-users <[email protected]> > [email protected] > > I think YCSB can handle that, but I am not sure about the 100% random part. > > On Wed, Sep 26, 2012 at 4:25 AM, Dalia Hassan <[email protected] > >wrote: > > > Hello, > > > > Could anyone help me how to measure Hbase random read performance on a > > cluster > > > > Please reply asap > > > > Thanks, > > > > > > -- > Kevin O'Dell > Customer Operations Engineer, Cloudera >
