code review 6561069: textproto: string interning on headers to reduce allocations

textproto: string interning on headers to reduce allocations

Saving one allocation per header for common HTTP
headers results in more transactions/GC in a
webserver and a lower "GC tax" per transaction,
and a higher throughput.

  benchmark                  old ns/op    new ns/op    delta
  BenchmarkReadMIMEHeader        12811        13240   +3.35%

  benchmark                 old allocs   new allocs  pct new/old
  BenchmarkReadMIMEHeader        32.00        24.00        75.00

Timings are noisy; worst case in 5 trials above; the average
is 0% time delta.

Also, updated benchcmp to show alloc savings.

[jeff.allen, 2012-09-28 10:25:21]

Hello bradfitz@golang.org (cc: golang-dev@googlegroups.com),

I'd like you to review this change to
https://code.google.com/p/go

[dave_cheney.net, 2012-09-28 12:16:17]

On 2012/09/28 10:25:21, jeff.allen wrote:
> Hello mailto:bradfitz@golang.org (cc: mailto:golang-dev@googlegroups.com),
> 
> I'd like you to review this change to
> https://code.google.com/p/go

Awesome stuff. Could I ask you to submit a second CL for the benchcmp change, so
it doesn't get mixed into this one?

[gburd, 2012-09-28 15:37:57]

http://codereview.appspot.com/6561069/diff/5001/src/pkg/net/textproto/reader.go
File src/pkg/net/textproto/reader.go (right):

http://codereview.appspot.com/6561069/diff/5001/src/pkg/net/textproto/reader....
src/pkg/net/textproto/reader.go:644: headerPool[hk] = h
The strings in commonHeaders are in the format commonly used, but they are not
canonical. Use

headerPool[hk] = CanonicalMIMEHeaderKey(h)

[jeff.allen, 2012-10-03 12:22:24]

*** Abandoned ***

[bradfitz, 2012-10-03 13:18:40]

Did you mean to abandon this? I was out Friday and Monday or I would've
replied sooner.
 On Oct 3, 2012 5:22 AM, <jeff.allen@gmail.com> wrote:

> *** Abandoned ***
>
>
http://codereview.appspot.com/**6561069/<http://codereview.appspot.com/6561069/>
>

[jeff.allen, 2012-10-03 13:55:26]

Was not totally on purpose (I clearly don't know the difference between -d and
-D), but I am working on a better measurement of the impact, and will send it in
again when that's done. Sorry for the noise.