Strange pattern of execution time

[Eugene Strokin]

Hello, 

we have a problem we cannot figure out for a long time.

We switched from Hector to Astyanax and before everything was looking stable, but now we see execution slower, but not just slower.

For example, we make 10 calls to Cassandra from our application server.

Response time of 1 of them in 0-10 ms range, as expected.

5 of them in 50-60 ms range,

and 4 of them in 100-110 ms range.

So we see 50ms step all the time. On a load test it goes up to 200-210 ms. If we switch back to Hector everything looks good.

It looks like, that there is 50ms time out for some reason and than retry.

Could anybody give us any kind of direction or idea? Because we spent already a lot of time, and still cannot understand the problem

Thanks,

Eugene