> There is a wonderful document in the kernel source tree,
> it is called linux/Documentation/networking/ip-sysctl.txt
>
> Please, do read it and search for tcp_mem, it will answer
> your questions.
Damned, forget my previous mail regarding bytes vs. pages, I misread.
However, I'd be glad if you could tell me why my writes block that long.
I'm certainly under 128MB of socket buffers, so the limits (on my
1GB machine) should not be reached.
Here's an strace excerpt:
08:03:05.986018 read(4, "GET /big HTTP/1.0\r\nHost: stars20"..., 4096) = 49 <0.000018>
08:03:05.986113 read(5, "GET /big HTTP/1.0\r\nHost: stars20"..., 4096) = 49 <0.000021>
08:03:05.986210 read(6, "GET /big HTTP/1.0\r\nHost: stars20"..., 4096) = 49 <0.000014>
08:03:05.986277 read(7, "GET /big HTTP/1.0\r\nHost: stars20"..., 4096) = 49 <0.000017>
08:03:05.986350 poll([{fd=3, events=POLLIN, revents=POLLIN}, {fd=4, events=POLLOUT, revents=POLLOUT}, {fd=5, events=POLLOUT, revents=POLLOUT}, {fd=6, events=POLLOUT, revents=POLLOUT}, {fd=7, events=POLLOUT, revents=POLLOUT}], 5, -1) = 5 <0.000019>
08:03:05.986453 accept(3, {sin_family=AF_INET, sin_port=htons(4618), sin_addr=inet_addr("62.104.210.74")}}, [16]) = 8 <0.000024>
08:03:05.986579 accept(3, 0xbffffc34, [16]) = -1 EAGAIN (Resource temporarily unavailable) <0.000014>
08:03:05.986639 write(4, "HTTP/1.0 200 OK\r\nDate: Tue, 29 A"..., 49342) = 49342 <0.048541>
08:03:06.035279 close(4) = 0 <0.000033>
08:03:06.035401 write(5, "HTTP/1.0 200 OK\r\nDate: Tue, 29 A"..., 49342) = 49342 <0.065298>
The writes _all_ look that way.
regards
Patrick
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
