[squid-users] Big problem with FD's in SUN E4000

Hi,
I have a Enterprise E250 with Solaris 8 and Squid 2.4 stable 1, recently
I changed this machine to a Enterprise E4000 with Solaris 8 and Squid
2.4 stable 1, I used the same configuration of E250 for compiling and
configure this machine, but this machine not work fine.
It follows below a part of the archive of log.

2001/05/21 15:35:07| Starting Squid Cache version 2.4.STABLE1 for
sparc-sun-solaris2.8...
2001/05/21 15:35:07| Process ID 7744
2001/05/21 15:35:07| With 32768 file descriptors available
2001/05/21 15:35:07| DNS Socket created on FD 4
2001/05/21 15:35:07| Adding nameserver 127.0.0.1 from /etc/resolv.conf
2001/05/21 15:35:08| User-Agent logging is disabled.
2001/05/21 15:35:08| Unlinkd pipe opened on FD 31
2001/05/21 15:35:08| Swap maxSize 10240000 KB, estimated 787692 objects
2001/05/21 15:35:08| Target number of buckets: 39384
2001/05/21 15:35:08| Using 65536 Store buckets
2001/05/21 15:35:08| Max Mem size: 512000 KB
2001/05/21 15:35:08| Max Swap size: 10240000 KB
2001/05/21 15:35:08| Rebuilding storage in /cache/A (CLEAN)
2001/05/21 15:35:08| Using Least Load store dir selection
2001/05/21 15:35:08| Loaded Icons.
2001/05/21 15:35:08| Accepting HTTP connections at 0.0.0.0, port 3128,
FD 33.
2001/05/21 15:35:08| Accepting ICP messages at 0.0.0.0, port 3130, FD 34.
2001/05/21 15:35:08| WCCP Disabled.
2001/05/21 15:35:08| Ready to serve requests.
2001/05/21 15:35:10| Store rebuilding is 10.3% complete
2001/05/21 15:35:12| Done reading /cache/A swaplog (39809 entries)
2001/05/21 15:35:12| Finished rebuilding storage from disk.
2001/05/21 15:35:12| 39809 Entries scanned
2001/05/21 15:35:12| 0 Invalid entries.
2001/05/21 15:35:12| 0 With invalid flags.
2001/05/21 15:35:12| 39809 Objects loaded.
2001/05/21 15:35:12| 0 Objects expired.
2001/05/21 15:35:12| 0 Objects cancelled.
2001/05/21 15:35:12| 0 Duplicate URLs purged.
2001/05/21 15:35:12| 0 Swapfile clashes avoided.
2001/05/21 15:35:12| Took 3.4 seconds (11718.1 objects/sec).
2001/05/21 15:35:12| Beginning Validation Procedure
2001/05/21 15:35:12| Completed Validation Procedure
2001/05/21 15:35:12| Validated 39809 Entries
2001/05/21 15:35:12| store_swap_size = 306439k
2001/05/21 15:35:12| storeLateRelease: released 0 objects
2001/05/21 15:38:08| comm_accept: FD 33: (130) Software caused
connection abort
2001/05/21 15:38:08| httpAccept: FD 33: accept failure: (130) Software
caused connection abort
2001/05/21 15:38:30| comm_accept: FD 33: (130) Software caused
connection abort
2001/05/21 15:38:30| httpAccept: FD 33: accept failure: (130) Software
caused connection abort
2001/05/21 15:38:32| comm_accept: FD 33: (130) Software caused
connection abort

These messages happen again it each 10 seconds approximately.
The machine E250 is more fast that the E4000 for any type of access
(ftp, http) !?!.
So I think that the problem is occurs when the limit of connections
exceeds the default value. I read the Solaris documentation (Tune) and I
increase this values in the tcp device:

tcp_conn_req_max_q 1024
tcp_conn_req_max_q0 8192
tcp_smallest_anon_port 8192
udp_smallest_anon_port 8192
tcp_xmit_hiwat 524280
tcp_recv_hiwat 524280
tcp_max_buf 8388608
tcp_cwnd_max 65534

Hardware Information:

Sun Enterprise E4000
SunOS proxy 5.8 Generic_108528-06 sun4u sparc SUNW,Ultra-Enterprise
2.5 Gb RAM
4 x SUNW,UltraSPARC-II 248Mhz
64 bit support
1 GigaBit Interface (ge0), setting up at 1000Mbps Full Duplex Forced

In the E250 this problem not occurs, and the compilation mode is the
same of the E4000. All modifycation in the Solaris (SO) I had recompiled
the squid.
Someone has any tip for this problem ?

I thank any help.

Sorry any errors in this email. :-)

Best Regards

_______________________________________
Hederson Velasco Ramos
Technology Consultant
EI - Network Services
Email: hederson.ramos@edb.ericsson.se
Received on Mon May 21 2001 - 16:25:10 MDT