Re: [squid-users] Solaris tuning

On Sunday 09 December 2001 11.48, Daniel Baldoni wrote:

> tcp_conn_req_max_q 1024 128
> tcp_conn_req_max_q0 2048 1024

Hmm.. hasn't these been addressed by SUN alrady as part of their SYN flood
protection?

> tcp_time_wait_interval 60000 240000

This should not really require tuning in modern OS:es.. and going too low
will cause some TCP/IP problems. Should not be less than 2 minutes.

> arp_cleanup_interval 60000 300000

Should not ever be needed unless your network is very poorly designed, why
tune it?

> ip_ignore_redirect 1 0
> ip_send_redirects 0 1
> ip_forward_src_routed 0 1
> ip_respond_to_echo_broadcast 0 1
> ip_respond_to_address_mask_broadcast 0 0
> ip_respond_to_timestamp_broadcast 0 1

All makes sense, and should be the factory default imho.

> throughput. Having said that, I notice that Squid calls setsockopt() with
> SO_RCVBUF (cancelling out any use of tcp_recv_hiwat) but I can't find any
> use of SO_SNDBUF - is it worth adjusting tcp_xmit_hiwat?

Probably only marginally.

> I'm very unsure about the potential impact some of the above timeouts might
> have on the proxies' respective performance levels.

Make sure the tunings you are reading is really intended for the Solaris
version you are running. SUN has improved their TCP/IP stacks quite a bit
over the years, and in many cases tunings that were appropriate on older
versions to work around some problem is no longer recommendable as SUN has
made a better solution to the problem the old tuning tries to address.

> Now, let's complicate things even further. Once I got stuck into the
> Tunables Manual, I kept on reading (always a bad thing <grin>). I am now
> wondering about the potential benefits (or pitfalls?) of reducing maxusers
> down to say, 64 (or maybe even lower).

There may mainly be some memory to save in tuning down things not needed by
Squid, but not very much I think.

Regards
Henrik

-- 
MARA Systems AB
Giving you basic free Squid support
Priority support or Squid enhancements available on request