On 1/4/2018 8:13 PM, tanjj1990 wrote:
> Sorry, I've upload my tests as attachments.

Please don't top post it makes the conversation hard to follow [1].

I've reviewed the test results and they show the same thing as what you
initially reported, except that the proxy tests w/ PostUpload do not
have a consistent value at 28,27,13,13 and 7 seconds. Probably there was
some congestion at that time, but that doesn't explain the overall
results which seem to be otherwise consistent.

You could try adjusting the socket send buffer size. Normally libcurl
skips adjusting it in Windows Vista and higher [2], but I would like to
see if there's a difference if you increase it. Can you try 16k, 32k,
and 64k? I have attached a patch that sets it to 64k.

[1]: https://curl.haxx.se/mail/etiquette.html#Do_Not_Top_Post
[2]: https://github.com/curl/curl/blob/curl-7_57_0/lib/connect.c#L944-L968

diff --git a/lib/connect.c b/lib/connect.c
index 3edb71e..d09ef01 100644
--- a/lib/connect.c
+++ b/lib/connect.c
@@ -943,7 +943,8 @@ static void nosigpipe(struct connectdata *conn,
 
 void Curl_sndbufset(curl_socket_t sockfd)
 {
-  int val = CURL_MAX_WRITE_SIZE + 32;
+  int val = 64*1024;
+/*
   int curval = 0;
   int curlen = sizeof(curval);
 
@@ -963,7 +964,7 @@ void Curl_sndbufset(curl_socket_t sockfd)
   if(getsockopt(sockfd, SOL_SOCKET, SO_SNDBUF, (char *)&curval, &curlen) == 0)
     if(curval > val)
       return;
-
+*/
   setsockopt(sockfd, SOL_SOCKET, SO_SNDBUF, (const char *)&val, sizeof(val));
 }
 #endif
-------------------------------------------------------------------
Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library
Etiquette:   https://curl.haxx.se/mail/etiquette.html

Reply via email to