nomi benchmark of fleet cluster

A set of benchmark results done in a multi-node fleet cluster via nomi.

Benchmark environment

• 1 main node : etcd1 (8 vCPUs, 12GB RAM), CentOS 7
• 3 main nodes : fleet1, fleet2, and fleet3 (each 2 vCPUs, 4GB RAM, CoreOS 1122)
• etcd v2.3.7 or v3.0.9
• fleet master (v0.13.0-234-g774ac312, 2016-11-23) or particular branches
• gRPC enabled

How to run nomi benchmarks

Set up multiple nodes in a cluster. Let's assume that we have 4 nodes.

Build fleet and copy binaries to each cluster. You can use copyfleet script.

$ cd $GOPATH/src/github.com/coreos/fleet
$ git checkout master
$ copyfleet master

Restart etcd and fleetd on each node, using stopall and startall.

$ stopall
... # make sure it was cleanly shut down, etc.
$ startall

Prepare nomi configs based on examples provided by nomi, and put them into nomi-config directory on the main node.

Run nomi benchmarks on the main node, using run-nomi.

$ ssh etcd1
# ./run-nomi

Configuration of nomi

The config benchmarkDef3000Units.yaml is like below:

instancegroup-size: 1
instructions:
  - sleep: 1
  - start:
     max: 3000
     interval: 100
  - sleep: 600
  - stop: stop-all

Note that you might need to change the sleep parameter after the start command according to the value of max. To increase max, you should increase also sleep. For example, to test 10000 units, you should set max to 10000, as well as increase sleep from 600 to 2000. Otherwise nomi could not retrieve all results.