Squid slowing conections

Hi all,

We are using squid at our ISP for a long time, without
any problems.

Recently, the squid starts to slow the internet conections:

        - high delay to get information from the URLs
        - it seems it is caching thinks before send it
          to the client
        
We put squid off and, even without cache, we get best results
for clients with dial-up conections.

Here some aspects of configuration:

- we are using transparent caching and the ASs are pointing to
  squid as the default router;

- the machine is a Pentium II 333 with 128 MB RAM and
  8GB EIDE just for squid;

- we have a maximum of 100 modem working at same time;

- squid conf:

        cache_mem 64 MB
        maximum_object_size 5120 KB
        ipcache_size 1024
        cache_dir /var/spool/squid 8000 32 256
        dns_children 15
        refresh_pattern -i \.gif$ 14400 80% 43200
        refresh_pattern -i \.tif?f$ 14400 80% 43200
        refresh_pattern -i \.bmp$ 14400 80% 43200
        refresh_pattern -i \.jpg$ 14400 80% 43200
        refresh_pattern -i \.jpeg$ 14400 80% 43200
        refresh_pattern -i \.xbm$ 14400 80% 43200
        refresh_pattern -i \.png$ 14400 80% 43200
        refresh_pattern -i \.wrl$ 14400 80% 43200
        refresh_pattern -i \.ico$ 14400 80% 43200
        refresh_pattern -i \.sit$ 14400 80% 43200
        refresh_pattern -i \.zip$ 14400 80% 43200
        refresh_pattern -i \.hqx$ 14400 80% 43200
        refresh_pattern -i \.exe$ 14400 80% 43200
        refresh_pattern -i \.arj$ 14400 80% 43200
        refresh_pattern -i \.lzh$ 14400 80% 43200
        refresh_pattern -i \.lha$ 14400 80% 43200
        (and a few others like these)
        refresh_pattern . 0 20% 4320
        reference_age 2 weeks
        quick_abort_min 16 KB
        quick_abort_max 16 KB
        quick_abort_pct 95
        httpd_accel_host virtual
        httpd_accel_port 80
        httpd_accel_with_proxy on
        delay_pools 2
        delay_class 1 1
        delay_class 2 2
        delay_access 1 allow src_lan (LAN IPs)
        delay_access 1 deny all
        delay_access 2 allow src_clientes (MODEMS IPs)
        delay_access 2 deny all
        delay_parameters 1 65536/65536
        delay_parameters 2 8192/16384 4096/4096
        delay_initial_bucket_level 50
        
Now, some questions I need your experience and help:

- Is there any modifications for the parametes above to increase
  squid performance with there configuration?

- If the MODEMS supports only 33,600 kbps, is the delay pools right
configured?

- If this machine is the gateway, is the problem with squid or with
IPFWADM
  to resend the packats for the real router in the network?

        IPFWADM configuration:
                ipfwadm -I -a a -P tcp -S any/0 -D 127.0.0.1 80
                ipfwadm -I -a a -P tcp -S any/0 -D squid_IP_addr 80
                ipfwadm -I -a a -P tcp -S any/0 -D any/0 80 -r 3128

- I could tell to squid not to cache some URL. What is the best way
(faster for
  squid) to to this?

- Waht parameters from cachemgr could I use to measure the eficiency of
  squid and to determine if the machine is enough for running the
  cache?

I hope you can tell me something about this and, if you need some
more information, please let me know.

TIA,

----------
George Amaro
Brazilian Agricultural Research Corporation
Network Engineering Team
Received on Wed Feb 23 2000 - 05:26:32 MST