Benchmarking etcd v2.2.0

GCE n1-highcpu-2 machine type

1x dedicated local SSD mounted as etcd data directory
1x dedicated slow disk for the OS
2x CPUs

3 etcd 2.2.0 members, each runs on a single machine.

Bootstrap another machine, outside of the etcd cluster, and run the HTTP benchmark tool with a connection reuse patch to send requests to each etcd cluster member. See the for the patch and the steps to reproduce our procedures.

The performance is calculated through results of 100 benchmark rounds.

Single Key Write Performance

key size in bytes	number of clients	target etcd server	average write QPS	write QPS stddev	average 90th Percentile Latency (ms)	latency stddev
64	1	leader only	55	4	24.51	13.26
64	64	leader only	2139	125	35.23	3.40
64	256	leader only	4581	581	70.53	10.22
256	1	leader only	56	4	22.37	4.33
256	64	leader only	2052	151	36.83	4.20
256	256	leader only	4442	560	71.59	10.03
64	64	all servers	1625	85	58.51	5.14
64	256	all servers	4461	298	89.47	36.48
256	64	all servers	1599	94	60.11	6.43
256	256	all servers	4315	193	88.98	7.01

Write QPS to cluster leaders seems to be increased by a small margin. This is because the main loop and entry apply loops were decoupled in the etcd raft logic, eliminating several blocks between them.
Write QPS to all members seems to be increased by a significant margin, because followers now receive the latest commit index sooner, and commit proposals more quickly.