Open berezovskyi-oleksandr opened 10 months ago
Thank you very much for your PR. Could you provide their experiental report?
In fact, chunk size is only an approximate value which always less than the preset value(jpg images are compressed). So should it be marked with approximate value on tip?
I observe significant improvement in results with 4k "size" for shitty linlks.
Ran a set of tests with tc
+netem
:
tc qdisc change dev eth0 root handle 1: netem delay 300ms 25ms loss 1%
Experiment 1:
Download | Upload | Latency | |
`speedtest-go` with 1k chunk | 6263587 B/s | 658662 B/s | 297 ms |
`speedtest-cli` | 26171990 B/s | 636274 B/s | 304 ms |
24% | 103% |
Experiment 2:
Download | Upload | Latency | |
`speedtest-go` with 4k chunk | 8072362 B/s | 747100 B/s | 296 ms |
`speedtest-cli` | 8929612 B/s | 637856 B/s | 300 ms |
90% | 117% |
On artificially degraded link with stable conditions, 4k version of speedtest-go
shows results significantly closer to those, shown by 1k: 10% vs 76% difference from the reference implementation
@r3inbowari are there any chances to get this merged?
are there any chances to get this merged?
Yes, we will merge it after vacation.
Hi, @berezovskyi-oleksandr, I make a change to fix multi server test error. could you test it, if ok, we can merge it now.
I'd like to be able to specify this as well.
It might be crucial to have the possibility to configure the chunk sizes. I observed that in some cases chunk size impacts the results.