Benchmarking etcd v2.2.0-rc
GCE n1-highcpu-2 machine type
- 1x dedicated local SSD mounted under /var/lib/etcd
- 1x dedicated slow disk for the OS
- 2x CPUs
3 etcd 2.2.0-rc members, each runs on a single machine.
Also, we use 3 etcd 2.1.0 alpha-stage members to form cluster to get base performance. etcd’s commit head is at , which is the same as the one that we use in etcd 2.1 benchmark.
Bootstrap another machine and use the to send requests to each etcd member. Check the benchmark hacking guide for detailed instructions.
key size in bytes | number of clients | target etcd server | write QPS | 90th Percentile Latency (ms) |
---|---|---|---|---|
64 | 1 | leader only | 76 (+22%) | 19.4 (-15%) |
64 | 64 | leader only | 2461 (+45%) | 31.8 (-32%) |
64 | 256 | leader only | 4275 (+1%) | 69.6 (-10%) |
256 | 1 | leader only | 64 (+20%) | 16.7 (-30%) |
256 | 64 | leader only | 2385 (+30%) | 31.5 (-19%) |
256 | 256 | leader only | 4353 (-3%) | 74.0 (+9%) |
64 | 64 | all servers | 2005 (+81%) | 49.8 (-55%) |
64 | 256 | all servers | 4868 (+35%) | 81.5 (-40%) |
256 | 64 | all servers | 1925 (+72%) | 47.7 (-59%) |
256 | 256 | all servers | 4975 (+36%) | 70.3 (-36%) |
write QPS to leader is increased by 20~30%. This is because we decouple raft main loop and entry apply loop, which avoids them blocking each other.
write QPS to all servers is increased by 30~80% because follower could receive latest commit index earlier and commit proposals faster.