[squid-users] accelerator accept problems

Hi

I'm using Squid in accelerator mode for a fairly busy Web site (ca.
200-250 req/s in peak). Everything is currently hosted on the same server
- Squid, Apache and MySQL, but the machine is powerful enough (2xXeon
3,2GHz, 2GB RAM) to handle the load (50% idle, no swap used, 1GB cached,
300MB for buffers). I have around 60% hit rate for requests - all static
content is cached in memory, and dynamic pages are forwarded to Apache.

I have a problem with Squid not accepting connecting in a timely manner.
Here is a tcpdump snippet:

16:38:37.240306 IP client.42054 > squid.80: S 487460662:487460662(0) win
5840 <mss 1460,nop,nop,timestamp 2904533217 0,nop,wscale 0>
16:38:40.239044 IP client.42054 > squid.80: S 487460662:487460662(0) win
5840 <mss 1460,nop,nop,timestamp 2904536217 0,nop,wscale 0>
16:38:46.237981 IP client.42054 > squid.80: S 487460662:487460662(0) win
5840 <mss 1460,nop,nop,timestamp 2904542217 0,nop,wscale 0>
16:38:58.233859 IP client.42054 > squid.80: S 487460662:487460662(0) win
5840 <mss 1460,nop,nop,timestamp 2904554217 0,nop,wscale 0>
16:38:58.233872 IP squid.80 > client.42054: S 83236182:83236182(0) ack
487460663 win 5792 <mss 1460,nop,nop,timestamp 402554534
2904554217,nop,wscale 0>
16:38:58.233973 IP client.42054 > squid.80: . ack 1 win 5840
<nop,nop,timestamp 2904554217 402554534>

client and squid are connected to the same Gigabit Ethernet switch, so
there is no problem with connectivity between them. What can cause such
behavior? Is it a problem with Squid itself, Squid configuration or maybe
kernel configuration?

I'm running Debian Linux sarge, kernel 2.6.8-2-686-smp with
/proc/sys/net/ipv4/tcp_max_syn_backlog=4096.

My squid.conf:

http_port ip1:80
http_port ip2:80
icp_port 0
hierarchy_stoplist cgi-bin ?
acl QUERY urlpath_regex cgi-bin
no_cache deny QUERY
cache_mem 256 MB
cache_dir ufs /var/spool/squid 512 16 256
cache_store_log none
debug_options ALL,1
hosts_file /etc/hosts
redirect_rewrites_host_header off
refresh_pattern . 0 20% 4320
read_timeout 30 seconds
request_timeout 30 seconds
persistent_request_timeout 10 seconds
half_closed_clients off
pconn_timeout 120 seconds
acl all src 0.0.0.0/0.0.0.0
acl manager proto cache_object
acl localhost src 127.0.0.1/255.255.255.255
acl to_localhost dst 127.0.0.0/8
acl purge method PURGE
acl CONNECT method CONNECT
http_access deny blokuj
http_access allow manager localhost
http_access deny manager
http_access allow purge localhost
http_access deny purge
http_access deny !Safe_ports
http_access deny CONNECT !SSL_ports
http_access allow all
http_reply_access allow all
icp_access deny all
httpd_accel_host 127.0.0.1
httpd_accel_port 80
httpd_accel_single_host on
httpd_accel_with_proxy off
httpd_accel_uses_host_header on
httpd_accel_no_pmtu_disc on
buffered_logs on
coredump_dir /var/spool/squid
client_persistent_connections off
server_persistent_connections on
pipeline_prefetch on

Any help would be greatly appreciated.

Best regards,
Michal Gruchala
Received on Wed Jul 20 2005 - 09:00:37 MDT