Re: DIRECT vs PARENT

On Mon, 18 Aug 1997, Stavros Patiniotis wrote:

> Hello,
> I have had squid installed for some time now, and was looking to
> better configure it for "policy based routing". My setup is as follows:
>
> PEERING INTERNATIONAL
> EXCHANGE
> | |
> | |(Proxy Cache)
> | 10Mbps 512K |
> |---------------+-----------------|
> |
> |
> ME
>
> Now the Peering Excahnge (PE) is a very low cost link, while the
> International (I) link is very expensive :(
>
> Now the problem comes into play as we *have to* set the "I" link's proxy
> server as a PARENT, if we don't, we pay 6c /Mb more for data. Also most
> data is delivered via the "I" link, rather than the PE. I have noticed
> the directives of Minimum_hop and Local_domain. A squid "cloud" is setup
> by each of the participants in the PE, ie we all have each other set as
> siblings (another question for later...).
>
> The problem is that any web sites located within the PE maybe (I say
> maybe as it doesn't seem to be always using the PARENT) fetched
> by the Parent, and we want to go DIRECT. The Local_domain directive does
> exactly what we need it to, ie go direct
> instead of via the parent, however this is very unscalble! The minimum
> hop directive also sounds good, as most webpages via the PE are within 5
> or so hops, where as the "I" data is about 8 hops away.
>
> I started to check some of the logs that were being created when I was
> accessing pages. I noticed that most, if not all pages fetched directly
> from the PE came back as DIRECT, even though the PINGER support has not
> been compiled in. I also tried another local ISP (not on the PE, but on
> the same router as me connected to the "I" provider) and their pages came
> DIRECT also! The only reason that I could deduce is that squid must
> automatically ping (EVEN *without* pinger support compiled in) the site
> to determine the RTT, and if it is less than the RTT of the parent, then
> go DIRECT. Just to confirm that the minimum_hop function was not active,
> I set it to 20 and kill -1 pid, and got the same results.
>
> QUESTIONS:
> Why is squid going direct on some pages when it has a parent set (the
> parent is alive)?
>
> Does a function such as minimum_hop work efficiently, or does it really
> slow squid down, remember these squids are getting 250,000 hits per day
> plus udp+tcp traffic from (currently) 4 other ISP's? Doing a traceroute
> to *each* site is very time consuming, and no doubt CPU intensive, how
> does this work?
>
> In future when this PE becomes national and international, the
> minimum_hop function will not work as the hops will have increased. Is
> their any intent to integrate squid with a routing protocol- such as BGP,
> ie do next hop lookups for each host (like the DNS lookups). ie have a
> directive such as: (For a dual homed provider)
>
> (NAME) (ASN) (METHOD)
> Fetch_Method ASXXXX DIRECT
> Fetch_Method ASYYYY PARENT
> As you get more internet gateways, add more Fetch_Method Directives...
>
> Where in the above (right at the top) diagram, ASXXXX would be the ASN
> for the PE, and ASYYYY would be the ASN for the "I" link.
>
> Your comments would be appreciated...
 
We are in the same situation. What we do is import the bgp tables of
our border router into squid "local_ip" statements ..
This is still not ideal, because local_ip caches the objects.

Regards
Peter Marelas

--
Phase One Interactive - Sun Solaris/Unix/Networking Consultant
P.O Box 549, Templestowe 3106 Melbourne, Australia
URL: http://www.phase-one.com.au/