Strange one-way sibling communication

    Hi all,

    I've searched high and low the Net and the archives for this list but did not found anything similar to the problem I'm getting. Maybe I just plain suck, but I'm trying your help as a last resort.

    In the company I work for (a small ISP) we run two independent services, both share the same link to our backbone (but one have a bandwidth limitation using FreeBSD) and both have proxy-caching with squid. Let's call "Service A" the one that don't have any bandwidth limitation, and "Service B" the slower, bandwidth-limited system. "Service A" and "Service B" networks are independent, but "Service B" network access the company's backbone-provider through "Service A" network. "Service B" proxy-cache machine also acts as the gateway between "Service B" network and "service A" network.

    THE PROBLEM: I configured both machines to be siblings (of each other), but apparently only "Service B" machine sends ICP requests to "Service A" machine ("Service A" machine doesn't send nothing to "Service B"). When using cachemgr.cgi, I always get something similar to the below in the "Peer Cache Statistics" section:
    
### cachemgr in MACHINE A (SERVICE A)
Sibling : machine_B_ip/3128/3130
Flags :
Address[0] : machine_B_ip
Status : Up
AVG RTT : 0 msec
LAST QUERY : 965071044 seconds ago
LAST REPLY : 965071044 seconds ago
PINGS SENT : 0
PINGS ACKED: 0 0%
FETCHES : 0 0%
IGNORED : 0 0%
Histogram of PINGS ACKED:
keep-alive ratio: 0%

### cachemgr in MACHINE B (SERVICE B)
Sibling : machine_A_ip/3128/3130
Flags :
Address[0] : machine_A_ip
Status : Up
AVG RTT : 30 msec
LAST QUERY : 425 seconds ago
LAST REPLY : 425 seconds ago
PINGS SENT : 68
PINGS ACKED: 68 100%
FETCHES : 3 4%
IGNORED : 2 3%
Histogram of PINGS ACKED:
         ICP_HIT : 3 4%
        ICP_MISS : 65 96%
keep-alive ratio: 100%
-----------------------------------
(machine_X_ip is the IP address of Machine "X")

    As you can see, "Machine A" apparently *never* sends pings nor ICP queries to "Machine B" and shows very big numbers in "LAST QUERY" and "LAST REPLY" (maybe because it never sends nothing and, thus, can't count correctly).

    Configuration files for both machines are nearly identical (I checked them using diff). Both machines are working without problems as "stand-alone" proxy-cache servers for months. I've tried, in the past, make then act as siblings, but I've got the problem and, now, I'm really decided to make them work together.

    Examining "cache.log" files from machines A and B, both say they configured their respective siblings correctly (for example, Machine A says correctly "Configuring Sibling machine_B_ip/3128/3130").

    Below I describe what I use in these two systems:
### MACHINE A (SERVICE A)
Linux RedHat 6.1
Squid 2.3.STABLE3-1 (RPM)
P 133 with 192MB of RAM
All-SCSI drives

### MACHINE B (SERVICE B)
FreeBSD 3.1
Squid 2.3.STABLE4 (source code)
PII 333 with 256MB of RAM
All-IDE drives

    I wish to thanks, in advance, for anyone that can help me. I want to apologize, too, for my bad English.

    Regards,
    Cesar A. Carvalho

-----------
Cesar Augusto Carvalho
cesar.carvalho@intergate.com.br
Herman at BrasIRC
Received on Mon Jul 31 2000 - 13:52:11 MDT