perf regression in master about axon HOT 9 CLOSED

Yikes, most likely the multipart stuff. Perhaps the buffering and looping of buffered buffers buffers buffer? lol

from axon.

hahahaha yeah, meh for now, we can look into that for the next release

from axon.

Got any feelings on net.Socket.ondata()? I messed around with that and FreeList for the message stuff.

Master:

pub bound
sending 1000 per tick
sending 1023 byte messages
  [401464 ops/s] [81601]^C

      min: 1792 ops/s
     mean: 268004 ops/s
   median: 388253 ops/s
    total: 81656 ops in 2.83s
  through: 261.72265625 mb/s

My "gofaster" branch

pub bound
sending 1000 per tick
sending 1023 byte messages
  [37477 ops/s] [99801]^C

      min: 1704 ops/s
     mean: 300978 ops/s
   median: 424153 ops/s
    total: 99917 ops in 2.556s
  through: 293.923828125 mb/s

from axon.

hmm that seems like something the JIT should take care of after it starts to optimize. I forget exactly what but I think the freelist was just for the http parsers since they were more expensive to construct for whatever reason

from axon.

im going to close this for now, we can still look into it but my initial benchmarks were reaallllly unrealistic anyway

from axon.

we'll have to update the docs though after the next release

from axon.

we might want to look at --trace-inlining too I see a few of our calls in there

from axon.

looks like it's not optimizing Socket#pack either which is pretty critical haha

from axon.

LOL yeah that is an important one. Sounds good!

from axon.