asubiotto commented on PR #38729:
URL: https://github.com/apache/arrow/pull/38729#issuecomment-1814111866
Added a commit with the benchmark and another commit to have writers release
buffers. It's not noticeable in the benchmark because we're probably writing
too little but it doesn't hurt:
```
goos: darwin
goarch: arm64
pkg: github.com/apache/arrow/go/v15/arrow/ipc
│ benchmain │ benchclose
│
│ sec/op │ sec/op vs base
│
IPC/Writer/codec=plain-12 8.370µ ± 5% 8.264µ ± 2% -1.25% (p=0.043
n=10)
IPC/Reader/codec=plain-12 11.41µ ± 4% 11.13µ ± 4% -2.43% (p=0.014
n=10)
IPC/Writer/codec=zstd-12 1.089m ± 12% 1.096m ± 8% ~ (p=0.912
n=10)
IPC/Reader/codec=zstd-12 38.22µ ± 3% 36.45µ ± 6% ~ (p=0.052
n=10)
IPC/Writer/codec=lz4-12 16.09µ ± 2% 15.99µ ± 5% ~ (p=0.436
n=10)
IPC/Reader/codec=lz4-12 40.79µ ± 4% 13.06µ ± 4% -67.98% (p=0.000
n=10)
geomean 37.10µ 30.26µ -18.44%
│ benchmain │ benchclose
│
│ B/op │ B/op vs base
│
IPC/Writer/codec=plain-12 8.394Ki ± 0% 8.394Ki ± 0% ~ (p=1.000
n=10) ¹
IPC/Reader/codec=plain-12 8.646Ki ± 0% 8.646Ki ± 0% ~ (p=1.000
n=10) ¹
IPC/Writer/codec=zstd-12 35.52Mi ± 0% 35.52Mi ± 0% ~ (p=0.810
n=10)
IPC/Reader/codec=zstd-12 25.63Ki ± 0% 25.63Ki ± 0% ~ (p=0.063
n=10)
IPC/Writer/codec=lz4-12 14.21Ki ± 1% 14.16Ki ± 0% -0.35% (p=0.043
n=10)
IPC/Reader/codec=lz4-12 266.53Ki ± 0% 10.55Ki ± 0% -96.04% (p=0.000
n=10)
geomean 79.69Ki 46.50Ki -41.66%
¹ all samples are equal
│ benchmain │ benchclose
│
│ allocs/op │ allocs/op vs base
│
IPC/Writer/codec=plain-12 113.0 ± 0% 113.0 ± 0% ~ (p=1.000 n=10) ¹
IPC/Reader/codec=plain-12 126.0 ± 0% 126.0 ± 0% ~ (p=1.000 n=10) ¹
IPC/Writer/codec=zstd-12 271.0 ± 0% 271.0 ± 0% ~ (p=1.000 n=10) ¹
IPC/Reader/codec=zstd-12 219.0 ± 0% 219.0 ± 0% ~ (p=1.000 n=10) ¹
IPC/Writer/codec=lz4-12 167.0 ± 0% 167.0 ± 0% ~ (p=1.000 n=10) ¹
IPC/Reader/codec=lz4-12 154.0 ± 0% 149.0 ± 0% -3.25% (p=0.000 n=10)
geomean 167.1 166.1 -0.55%
¹ all samples are equal
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
