[prev in list] [next in list] [prev in thread] [next in thread]
List: linux-net
Subject: Slow Connections After Exceeding Ephemeral Port Size
From: Kyle Brandt <kyle () kbrandt ! com>
Date: 2010-05-22 22:46:41
Message-ID: AANLkTilj6cCsE_rJhfRx1uci_xrxqKRtROypg2m4P0Aw () mail ! gmail ! com
[Download RAW message or body]
Hi Everyone,
According to linux-2.6/Documentation/networking/ip-sysctl.txt:
"""
With tcp_tw_recycle enabled
(i.e. by default) range 1024-4999 is enough to issue up to
2000 connections per second to systems supporting timestamps
"""
However, my script is unable to send requests at a fast rate not to
long after my ephemeral range is exceeded (without reuse and recycle
is drops right when the range is exceeded). The application is
creating about 1000 connections a second to a server using a pool of
source ips. Eventually the connections (hits) will jump up again.
Watching netstat -tapn | wc -l on both the client and the server never
gets very high. When the hits on the client drops to zero, I see
netstat at 0 or 1 on the server.
The server is nginx in front of Apache, but I see no errors in either
log. I posted some output from my script below as well some sysctl
settings. Anyone have any ideas on how I might sustain around 1000
hits a second from the client with my script?
Thank you,
Kyle
Total Hits: 29287, Roughly 944 hits a second, ip queue is length 1249
Total Hits: 30233, Roughly 946 hits a second, ip queue is length 1249
Total Hits: 31186, Roughly 953 hits a second, ip queue is length 1250
Total Hits: 32130, Roughly 944 hits a second, ip queue is length 1249
Total Hits: 33076, Roughly 946 hits a second, ip queue is length 1249
Total Hits: 34027, Roughly 951 hits a second, ip queue is length 1249
Total Hits: 34976, Roughly 949 hits a second, ip queue is length 1249
Total Hits: 35921, Roughly 945 hits a second, ip queue is length 1249
Total Hits: 36859, Roughly 938 hits a second, ip queue is length 1249
Total Hits: 37383, Roughly 524 hits a second, ip queue is length 1249
Total Hits: 37383, Roughly 0 hits a second, ip queue is length 1249
Total Hits: 37383, Roughly 0 hits a second, ip queue is length 1249
Total Hits: 37389, Roughly 6 hits a second, ip queue is length 1249
Total Hits: 37389, Roughly 0 hits a second, ip queue is length 1249
Total Hits: 37389, Roughly 0 hits a second, ip queue is length 1249
Total Hits: 37400, Roughly 11 hits a second, ip queue is length 1249
Total Hits: 37400, Roughly 0 hits a second, ip queue is length 1249
cat /proc/sys/net/ipv4/ip_local_port_range
32768 61000
The following is set on both the server and the client:
cat /proc/sys/net/ipv4/tcp_tw_re*
1
1
cat /proc/sys/net/ipv4/tcp_timestamps
1
(Also tried low fin timeout on client and server (5) but still the same issue).
--
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic