Hi Yuji, Were both tests run from the same machine at close the same time? If not, noisy neighbors may be affecting your performance on different AWS instances.
You should verify that you're using the same version of Java during both tests. Also, ensure that you're using the same test instance (that is not running Cassandra) to connect to both Cassandra clusters. Cheers, Joaquin Joaquin Casares Consultant Austin, TX Apache Cassandra Consulting http://www.thelastpickle.com On Wed, May 10, 2017 at 5:01 AM, Yuji Ito <y...@imagine-orb.com> wrote: > Hi all, > > I'm trying a simple performance test. > The test requests select operations (CL.SERIAL or CL.QUORUM) by increasing > the number of threads. > There is the difference of the performance between C* installed by yum and > C* which I built by myself. > What causes the difference? > > I use C* 2.2.8. > One of them was installed by yum (# yum install cassandra22). > Another was acquired by git from https://github.com/apache/cass > andra/tree/cassandra-2.2.8 and built it by myself. > I changed cassandra.yaml to set `commitlog_sync: batch` and > `commitlog_sync_batch_window_in_ms: 2`. > > My environment: > - a cluster has 3 nodes > - node: AWS EC2 m4.large with 200 IOPS EBS volume > - Replication Factor: 3 > - 10000 rows > > Result: > ** yum > ==== select (CL.SERIAL) ==== > threads operations/sec > 1 188 > 2 156 > 4 434 > 8 396 > 16 837 > 32 1176 > 64 2206 > 128 4115 > 256 7272 > > ** git > ==== select (CL.SERIAL) ==== > threads operations/sec > 1 192 > 2 162 > 4 264 > 8 446 > 16 733 > 32 1114 > 64 1715 > 128 2776 > 256 3920 > > ** yum > ==== select (CL.QUORUM) ==== > threads operations/sec > 1 434 > 2 909 > 4 1481 > 8 1904 > 16 2666 > 32 3106 > 64 3555 > 128 5000 > 256 9014 > > ** git > ==== select (CL.QUORUM) ==== > threads operations/sec > 1 666 > 2 1538 > 4 2500 > 8 3333 > 16 4210 > 32 5333 > 64 6597 > 128 7356 > 256 8075 > > Thanks, > Yuji > >