squid+wccp No buffer space available

[Ventsislav Velkov]

Hi all,
I have a problem when trying to run my squid-cache with wccp.
The problem is that after running for 1 hour the network hangs .
I have in the message log "/kernel: xl0: watchdog timeout" and in the squid logs " No buffer space available"
The system is dual xeon @500Mhz with 2.5GB of RAM and 10x9.1GB SCSI,
FreeBSD 4.9-PRERELEASE from Thu Sep 11 and squid-2.5.STABLE3 build from ports.
My kernel:
maxusers        256
options         MSGMNB=16384    # max # of bytes in a queue
options         MSGMNI=512      # number of message queue identifiers
options         MSGSEG=8192     # number of message segments
options         MSGSSZ=128      # size of a message segment
options         MSGTQL=512      # max messages in system
options         SHMSEG=16
options         SHMMNI=32
options         SHMMAX=2097152
options         SHMALL=4096

options         MAXDSIZ="(2048UL*1024*1024)"
options         MAXSSIZ="(2048UL*1024*1024)"
options         DFLDSIZ="(2048UL*1024*1024)"

options         NMBCLUSTERS=65535
options         NBUF=16384

When I am port-forwarding the traffic to the squid without wccp the machine is stable and works fine.
The problem appears after setting squid with wccp and bringing gre interfaces.
I've tried different settings but with no effect. The squid passes arround 20 mbits/s 
with over 100 req/s.

Will appreciate all advices.

Best regards,

----------------------------------
Ventsislav Velkov
Network Administrator
Evrocom.Net
+359 32 607321