Re: wget | Draft: Initial test for dynamic buffer size for downloads (!32)

[@rockdaboot]

Tim Rühsen commented:


I did two runs with stock Debian wget against a 20GB file hosted by apache2 
locally. The bottleneck is apache2.
```
$ wget http://localhost/file-20g -O /dev/null
real    0m9,473s
user    0m1,131s
sys     0m4,644s

real    0m9,530s
user    0m1,091s
sys     0m4,971s
```

and built wget with -O2 from your branch and did the same:
```
real    0m9,351s
user    0m0,315s
sys     0m2,157s

real    0m9,293s
user    0m0,407s
sys     0m2,077s
```

So the CPU dropped by ~50%, while the wall time roughly stayed the same 
(apache2 bottleneck being at 100% CPU), throughput at 2.1 GB/s.
This is quite a good result :+1:

Can you add some metrics to see the actual increase in buffer size ?
I'll take closer look into the source code later (hopefully this weekend) and 
would like to make tests with real downloads as well.
Should we try to figure out some better optimization flags than just -O2 ?

-- 
Reply to this email directly or view it on GitLab: 
https://gitlab.com/gnuwget/wget/-/merge_requests/32#note_971482183
You're receiving this email because of your account on gitlab.com.