Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/badboy/rust-io-new-vs-old
new io vs. old io Rust benchmark
https://github.com/badboy/rust-io-new-vs-old
Last synced: 24 days ago
JSON representation
new io vs. old io Rust benchmark
- Host: GitHub
- URL: https://github.com/badboy/rust-io-new-vs-old
- Owner: badboy
- Created: 2015-03-22T14:05:51.000Z (almost 10 years ago)
- Default Branch: master
- Last Pushed: 2015-03-22T14:49:35.000Z (almost 10 years ago)
- Last Synced: 2024-12-24T02:01:34.882Z (28 days ago)
- Language: Rust
- Size: 1.73 MB
- Stars: 1
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Rust benchmark: old io vs. new io
While porting [rdb-rs](https://github.com/badboy/rdb-rs) I came across huge performance drops when switching over to the [new io module](http://doc.rust-lang.org/nightly/std/io/index.html).
This repository contains a small test case with the [old](old-io.rs) and [new](new-io.rs) module in use.
I know my implementation of `read_exact` is far from perfect or optimized and I wonder how to do it properly.
Update after short discussion in the IRC channel:
Thanks to the [capnproto-rust](https://github.com/dwrensha/capnproto-rust/blob/6c755f3bf3f60d6b88dda5d30fb11b4ac4b3b8c2/src/serialize.rs) implementation by @dwrensha this is now as fast as the old one (and hopefully correct as well).
Compile with:
```
make
# For unoptimized build:
make FLAGS=
```
Run performance test (requires Linux perf):
```
make perf
```
## Local results for the included `40mb.rdb`
```
$ rustc --version
rustc 1.0.0-nightly (e2fa53e59 2015-03-20) (built 2015-03-21)
$ make perf
perf stat ./new-io 40mb.rdb 10
reading file '40mb.rdb'
len: 10
total read: 6789100
Error occured: Error { repr: Custom(Custom { kind: Other, desc: "Premature EOF", detail: None }) }
Performance counter stats for './new-io 40mb.rdb 10':
51,792310 task-clock (msec) # 0,988 CPUs utilized
0 context-switches # 0,000 K/sec
1 cpu-migrations # 0,019 K/sec
91 page-faults # 0,002 M/sec
76.092.295 cycles # 1,469 GHz
17.007.436 stalled-cycles-frontend # 22,35% frontend cycles idle
4.666.648 stalled-cycles-backend # 6,13% backend cycles idle
198.183.724 instructions # 2,60 insns per cycle
# 0,09 stalled cycles per insn
40.159.421 branches # 775,394 M/sec
23.136 branch-misses # 0,06% of all branches
0,052395574 seconds time elapsed
perf stat ./old-io 40mb.rdb 10
reading file '40mb.rdb'
len: 10
total read: 6789100
File EOF
Performance counter stats for './old-io 40mb.rdb 10':
38,478493 task-clock (msec) # 0,990 CPUs utilized
0 context-switches # 0,000 K/sec
1 cpu-migrations # 0,026 K/sec
92 page-faults # 0,002 M/sec
73.208.076 cycles # 1,903 GHz
15.416.222 stalled-cycles-frontend # 21,06% frontend cycles idle
3.199.782 stalled-cycles-backend # 4,37% backend cycles idle
191.857.108 instructions # 2,62 insns per cycle
# 0,08 stalled cycles per insn
40.252.577 branches # 1046,106 M/sec
22.051 branch-misses # 0,05% of all branches
0,038877979 seconds time elapsed
```
Unoptimized build:
```
$ make -B FLAGS=
$ make perf
perf stat ./new-io 40mb.rdb 10
reading file '40mb.rdb'
len: 10
total read: 6789100
Error occured: Error { repr: Custom(Custom { kind: Other, desc: "Premature EOF", detail: None }) }
Performance counter stats for './new-io 40mb.rdb 10':
860,784463 task-clock (msec) # 0,999 CPUs utilized
12 context-switches # 0,014 K/sec
15 cpu-migrations # 0,017 K/sec
93 page-faults # 0,108 K/sec
2.058.030.887 cycles # 2,391 GHz
877.873.779 stalled-cycles-frontend # 42,66% frontend cycles idle
85.669.427 stalled-cycles-backend # 4,16% backend cycles idle
3.466.656.663 instructions # 1,68 insns per cycle
# 0,25 stalled cycles per insn
710.519.986 branches # 825,433 M/sec
708.532 branch-misses # 0,10% of all branches
0,861473071 seconds time elapsed
perf stat ./old-io 40mb.rdb 10
reading file '40mb.rdb'
len: 10
total read: 6789100
File EOF
Performance counter stats for './old-io 40mb.rdb 10':
674,233347 task-clock (msec) # 0,999 CPUs utilized
28 context-switches # 0,042 K/sec
4 cpu-migrations # 0,006 K/sec
93 page-faults # 0,138 K/sec
1.678.672.215 cycles # 2,490 GHz
647.754.789 stalled-cycles-frontend # 38,59% frontend cycles idle
75.167.751 stalled-cycles-backend # 4,48% backend cycles idle
2.832.107.349 instructions # 1,69 insns per cycle
# 0,23 stalled cycles per insn
592.338.044 branches # 878,536 M/sec
709.566 branch-misses # 0,12% of all branches
0,674874353 seconds time elapsed
```
Results on a 416 Mb file are [in a Gist](https://gist.github.com/badboy/24d4ba25679afd1b576d).