there is a system test in ZK that allows measuring latency and throughput. more details here: https://issues.apache.org/jira/browse/ZOOKEEPER-2023
On Mon, Nov 10, 2014 at 8:14 PM, Ibrahim <i.s.el-san...@newcastle.ac.uk> wrote: > > Hi folks > I am benchmarking the performance of Zookeeper using five servers. I am > just > testing the write workload (CREATE node) in both mode (sync and async). The > benchmark code is designed to measure the performance of a ZooKeeper > ensemble for a predetermined length of time (e.g., sustained handling of > create requests for 5 minutes seconds). > > It seems easy to measure the latency and throughput in Sync mode as there > is > not any process run in the background. > > Measuring latency Using Async mode can be achieved by stat command (four > word command). However, I am struggling to measure the throughput in Async > mode because when the 5 minutes elapsed, the requests are still processing > in Zookeeper and after some time will stop. Also when I try to use the stat > command I see the following output: > Latency min/avg/max: 1/375/1259 > Received: 201870 > Sent: 201619 > Connections: 1 > Outstanding: 0 > Zxid: 0x1000ff2c7 > Mode: follower > Node count: 1032597 > > In the above output there is Node count, this can be expressed as total > number of nodes created, but the finished time is unknown. I think it is > not > true to say the finished time is 5 minutes, because the processing request > is not finished when 5 minutes are elapsed because of Async mode. > > Can you share knowledge? > How can I solve this problem? > Can anyone advise me for best strategy of testing and measuring the > performance of write workload in Zookeeper? > > Ibrahim > > > > > -- > View this message in context: > http://zookeeper-user.578899.n2.nabble.com/Best-strategy-to-test-Zookeeper-tp7580580.html > Sent from the zookeeper-user mailing list archive at Nabble.com. >
