If you post the code you're using to test, it would be helpful. You should also use cassandra-stress to see if you get similar results.
On Thu, Jul 2, 2015 at 1:39 AM Serega Sheypak <[email protected]> wrote: > What is the reason to do that? I understand BatchStatement as a kind of > atomic insert hack. > How it can help me to solve concurrency problem? 1 thread with sync insert > gives me 1K ops/sec. 10 threads give me 20 ops/sec :) > > Here are metrics for single thread async insert: > > -- Timers > ---------------------------------------------------------------------- > > insertTimer > > count = 4641205 > > mean rate = 14971.58 calls/second > > 1-minute rate = 18410.90 calls/second > > 5-minute rate = 10555.73 calls/second > > 15-minute rate = 4511.58 calls/second > > min = 0.00 milliseconds > > max = 0.12 milliseconds > > mean = 0.01 milliseconds > > stddev = 0.01 milliseconds > > median = 0.01 milliseconds > > 75% <= 0.01 milliseconds > > 95% <= 0.01 milliseconds > > 98% <= 0.02 milliseconds > > 99% <= 0.02 milliseconds > > 99.9% <= 0.12 milliseconds > > > what should I do to reach better performance when i use several threads? > > 2015-07-02 10:34 GMT+02:00 Vova Shelgunov <[email protected]>: > >> Did you tried to use BatchStatement? >> On Jul 2, 2015 11:00 AM, "Serega Sheypak" <[email protected]> >> wrote: >> >>> Hi, I have weird driver behaviour. Can you help me please to find the >>> problem? >>> Problem: I try to insert data using 10 threads. >>> I see that 10 thread starts, they start to insert some data and then >>> they hung. It takes enormous amount of time to insert (seconds for 1K >>> inserts). It runs 1K per second if I use single thread to insert. >>> >>> Here is my code: >>> https://gist.github.com/seregasheypak/6ddf5b24cd1c195f5355 >>> >> >
