TCP Drain Test

This is an attempt to confirm or deny this issue with closing http response bodies.

These are my test results so far:

$ go version
go version go1.6 darwin/amd64

$ PORT=5099 go run main.go 10
Without drain (x10): 2.989648ms
   With drain (x10): 784.782µs

$ PORT=5099 go run main.go 100
Without drain (x100): 12.047307ms
   With drain (x100): 6.030919ms

$ PORT=5099 go run main.go 1000
Without drain (x1000): 65.449043ms
   With drain (x1000): 55.154988ms

$ PORT=5099 go run main.go 10000
Without drain (x10000): 581.107868ms
   With drain (x10000): 571.955831ms

$ PORT=5099 go run main.go 100000
Without drain (x100000): 5.692069876s
   With drain (x100000): 5.648899366s

UPDATE

Thanks to @shurcooL's comment I updated this example to use a TLS connection, which is really where this problem is illuminated. TLS connections are much more expensive to reopen and therefore gain more benefit from reuse.

I've updated my example to just hit Github's api over https:

$ go version
go version go1.6 darwin/amd64

$ go run main.go https://api.github.com/repos/google/go-github 10
Without drain (x10): 1.302103099s
   With drain (x10): 282.559111ms

$ go run main.go https://api.github.com/repos/google/go-github 100
Without drain (x100): 8.339789738s
   With drain (x100): 2.694648312s

$ go run main.go https://api.github.com/repos/google/go-github 1000
Without drain (x1000): 1m16.253147895s
   With drain (x1000): 32.260955511s