I am running some ssbench tests in a cluster with 10 object servers and 3 proxy servers.

The object size is 4KB and only 600 objects are there.

When I run the tests for 2 minutes, first 70 seconds it works properly then suddenly it slows down.

I send 50-50 read/write requests. After slowing down, proxy error log contains records saying Object PUT returning 503, 1/3 required connections and ERROR with Object server 172.31.0.144:6000/sdb1 re: Expect: 100-continue .

Why this behaviour occur?

With larger object sizes it shows periodic drops in throughput and stable performance in the middle of them?

Note: I checked the disk, it is not saturated.