On Fri, 1 Dec 2023, Dmitry Karpov via curl-library wrote:
I tried the latest master branch with Daniel's fix, and I still see large
deviations (up to 20%) from the speed limit set by
CURLOPT_MAX_RECV_SPEED_LARGE. The deviations depend on a network speed and
they get bigger when the speed limit gets far from the network speed.
I can't reproduce that. I do this to test:
$ curl --limit-rate 5M localhost/100G -o /dev/null
Here's my debug patch to verify the math in Curl_pgrsLimitWaitTime:
diff --git a/lib/progress.c b/lib/progress.c
index e96cbf7af..8473cfc07 100644
--- a/lib/progress.c
+++ b/lib/progress.c
@@ -298,19 +298,26 @@ timediff_t Curl_pgrsLimitWaitTime(curl_off_t cursize,
minimum *= 1000;
else
minimum = TIMEDIFF_T_MAX;
}
+ fprintf(stderr, "size % " CURL_FORMAT_CURL_OFF_T
+ " limit: %" CURL_FORMAT_CURL_OFF_T
+ " min: %ld\n",
+ size, limit, minimum);
+
/*
* 'actual' is the time in milliseconds it took to actually download the
* last 'size' bytes.
*/
actual = Curl_timediff_ceil(now, start);
if(actual < minimum) {
/* if it downloaded the data faster than the limit, make it wait the
difference */
- return (minimum - actual);
+ timediff_t delta = (minimum - actual);
+ fprintf(stderr, "took %ldms wait %ldms more\n", actual, delta);
+ return delta;
}
return 0;
}
--
/ daniel.haxx.se
| Commercial curl support up to 24x7 is available!
| Private help, bug fixes, support, ports, new features
| https://curl.se/support.html
--
Unsubscribe: https://lists.haxx.se/mailman/listinfo/curl-library
Etiquette: https://curl.se/mail/etiquette.html
