Marc,
Is it possible that a single legitimate transfer from a user
saturated the link and slowed down other connections? For example,
somebody could be downloading a huge file or playing streaming media
through your proxy. If that is the case, you may want to look at delay
pools. The pools are designed to limit the bandwidth a single user (or a
group of users) can consume at a time. See delay pools description in
squid.conf and FAQ for details. (Note that negative byte hit ratios
are not necessarily related to this problem)
If nobody on the list has a ready-to-use solution (I do not),
you may have to enable access logging to see what your Squid is actually
doing during when it is using all available bandwidth.
Alex.
P.S. Please do not post from <info@talent.com.au> address -- people get
spammed with auto-generated advertisement messages when replying to your
emails...
On Mon, 18 Oct 1999, Marc Lucke wrote:
> * No evidence, I don't keep the access logs, I send them to
> /dev/null. I have been running Squid/Linux for a long time with not
> so many problems.
> * I graph Squid's SNMP variables using MRTG & a formula that Henrick
> Nordstrom gave me when I was asking how to do such things. Normally
> the traffic coming in is less than or equal to the traffic going out
> but on both occassions MRTG showed Squid downloading at the maximum
> rate of our bandwidth unrelentlessly. Some clients reported that
> their downloads were down to 300 k (where normally they are getting
> about 5.5 K) and all other evidence available points to Squid (ie none
> of the other machines showed corelating traffic).
>
> I fully accept that this may not be a difficulty of Squid but I would
> like some suggestions, if the advanced users on this list are feeling
> magnaminous enough, as to what might explain the behaviour &/or if
> anyone else has had the behaviour too.
>
> Thanks Alex,
> Cheers
> Marc :-)
Received on Sun Oct 17 1999 - 21:01:42 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:48:56 MST