performance

README

Performance tests for dsync

This directory contains code for doing performance measurements for dsync.

You can either run it locally in a couple of terminals, or for a true test you can run it in the cloud on a bunch of servers. Depending on what you want to do you will need to configure (and build) the program differently since it needs the IP addresses for all the servers (nodes) participating in the locking process.

Building

$ cd performance
$ go build

Running on localhost

For a local test, you need to change the nodes array in performance.go to something like this (example for 4 terminal windows)

var nodes = []string{
	"127.0.0.1:12345",
	"127.0.0.1:12346",
	"127.0.0.1:12347",
	"127.0.0.1:12348"}

and go build, and then run the following in the first terminal

$ ./performance -p 12345

followed by ./performance -p 12346 in the second terminal, etc. all the way through to ./performance -p 12348 for the last terminal.

Note that the actual test only starts once the program is started in all terminals.

Running in the cloud

To measure the actual performance you will want to run it on a few servers in the cloud or on local servers. Again, you will need to change the nodes array in performance.go to something like the following with the actual IP addresses of the servers (imagine 8 servers)

var nodes = []string{
	"10.x0.y0.z0:12345",
	"10.x1.y1.z1:12346",
	"10.x2.y2.z2:12347",
	"10.x3.y3.z3:12348",
	"10.x4.y4.z4:12349",
	"10.x5.y5.z5:12350",
	"10.x6.y6.z6:12351",
	"10.x7.y7.z7:12352"}

Either build the performance program on each server or copy it over, and then run ./performance -p 12345 on the first server, ./performance -p 12346 on the second, etc. through to the last server. (Note that for this case you can set the port number to the same value.)

Performance

See here for the collective results.

Tweaking

By changing the number of parallel loops to get locks (parallel := 5) you can influence the overall CPU load.