Re: [squid-users] Is there a delay-pool FAQ someplace

From: Robert Collins <robertc@dont-contact.us>
Date: 24 Jul 2003 07:55:36 +1000

On Thu, 2003-07-24 at 00:40, Josh Kuperman wrote:

> Lets say there are three groups on three subnets:
>
> Research on 192.168.2.0/24
> Marketing on 192.168.3.0/24
> Administration on 192.168.1.0/24
>
> Scenario One: insuring an even distribution of available
> bandwidth. Let's say I have a 1Mbps interent connection and I know I
> need 100kbps for external connection, email, web server,etc on my
> network. That leaves 300kbps for each of the tree groups.Can I enforce
> that with delay pools.

Yes, with one class one pool per group.

> Scenario Two: insuring a hierarchical preference so the following
> types of documents are treated as more important:
>
> 1. Text, html, pdf,
> 2. graphics.
> 3. audio files
> 4. video files
> 5. anything being streamed.
>
> Nothing should be prevented per se, but the simpler and older basic
> web services should always have a good response, while streaming video
> and any bandwidth hogs I know about should be limited.

You can do this, but not at the same time as the 300kbps limit for the
groups above. What you can do to come close though is use a class 2
pool. The aggregate field covers the group, and the per-last-octect
field covers each user. Now, limit each user to some fraction of the
group - say 100kbps. That means that sustained traffic from a single
machine will not saturate that group's allowance on it's own.

Long term, delays pools could be enhanced to do the sort of heirarchical
logic you talk about there - in squid 3.0 that is (relatively) easy to
accomplish.

> Scenario three: Keeping available bandwidth proportionate. Say
> Marketing sub net has 10 machines. First, how can I make sure that the same
> percent of the Marketing groups bandwidth is available to each
> machine. Second, is there a way to keep it proportionate but not count
> machines when they aren't being used.

This is the default behaviour for squid - no machine is given preference
to resources. It's not strictly enforced - it depends on the origin
servers bandwidth. Again, use of a class 1 or 2 pool will likely do what
you need.

> Scenario four: Limit bandwidth based on the online service. For
> example say you know a lot of people will be using Aol Instant
> Messenger, which is fine from time to time. But as it is a text based
> protocol you know that should never really take any bandwidth and you
> know you can restrain it quite a bit before you render it
> unusable. How would you do that.

This is a variation on Secnario 2, and my answer there applies here as
well.

Cheers,
Rob

-- 
GPG key available at: <http://members.aardvark.net.au/lifeless/keys.txt>.

Received on Wed Jul 23 2003 - 15:55:58 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:18:15 MST