RE: Bandwidth and load sharing

Where, this solution is a partial solution, because you will have two
default gateways to the internet on the same machine, and it will pickup
only one of those two. So you will not get any balance.
You can however fill the routing tables, either from the routers if you have
any BGP peer with your providers or by hand, using a thumb of hand
algorithm, I mean choosing some networks for one interface and others for
the second one and entering those by hand. Some has used this in the past,
masking the net ip's just with 255.0.0.0 so only 255 entries will be there.

Daniel E Visbal

-----Original Message-----
From: Scott Hess [mailto:scott@avantgo.com]
Sent: Friday, October 01, 1999 11:10 AM
To: Daniel E. Visbal; 'Troy Settle'; Squid Users
Subject: Re: Bandwidth and load sharing

Couldn't you run three squids on one machine, and set the port numbers for
each differently? This assumes that you can make the port numbers work,
and also that you can tell two of the squids to bind to specific IP
addresses (looks like tcp_outgoing_address to me).

Actually, you could also probably do it on one machine with a pair of IP
aliases. use tcp_incoming_address to bind the proxy and the two parents to
the primary and two IP aliases, use tcp_outgoing_address to bind the
parents to their aliases, and use udp_incoming_address and
udp_outgoing_address to make them all talk together nicely.

In either case, it's probably important to make certain that none of the
proxies caches an item from another proxy's cache. If that happens, then
you're storing 3x the data and doing 3x the seeks. If you can prevent the
duplicate storage, you should end up with the approximate disk accesses
that you'd have had with a single proxy (though cached info will obviously
be stored all over the place unless you configure the parents to not cache
at all). This assumes that you don't need the scaled-up disk access
performance of having two caches.

Later,
scott

----- Original Message -----
From: Daniel E. Visbal <daniel.visbal@sta.sistecol.com>
To: 'Troy Settle' <st@i-plus.net>; Squid Users <squid-users@ircache.net>
Sent: Thursday, September 30, 1999 2:53 PM
Subject: RE: Bandwidth and load sharing

> I will like to hear about other solutions, using less computers.
>
> The solution I was able to come with was not with 2 but 3 computers since
> sibling is valid only for data on the cache.
> I do have one computer for the users proxy that have two parents, each
one
> to the different internet providers.
>
> On the computer that is being used as the proxy, the relationship with
the
> other proxies is setup as parent with closest-only option, and also has
ICMP
> enable during configure.
>
> cache_peer proxy1 parent 3128 3130 closest-only
> cache_peer proxy2 parent 3128 3130 closest-only
>
> This will give some balance bandwidth since it will retrieve the data
from
> the best possible connection at the time the request arrive.
>
> Daniel E Visbal
>
> -----Original Message-----
> From: Troy Settle [mailto:st@i-plus.net]
> Sent: Thursday, September 30, 1999 12:24 PM
> To: Squid Users
> Subject: Bandwidth and load sharing
>
>
>
> Hey all,
>
> We're in the process of evaluating Squid as a transparent proxy server,
and
> all looks good so far. But, there's one major problem we'll be facing.
>
> Currently, we're padding some of our BGP route announcements to balance
out
> our incoming bandwidth. About 1/2 of our dialup ports are padded, and
the
> other half are not.
>
> If I bind 2 IP addresses to the ethernet interface on our squid box, can
> Squid be configured to balance it's usage between those 2 addresses? If
> not, can anyone offer any suggestions for balancing out our bandwidht
usage
> once we deploy Squid?
>
> One thought that comes to mind is to have 2 caches, one on each subnet,
> configured as siblings. The only disadvantage to this, is that it would
> double our hardware costs.
>
> How are other ISPs with 4k users and 2 T1s handling this situation?
>
>
> TIA,
>
> Troy Settle
> iPlus Internet Services
>
Received on Fri Oct 01 1999 - 11:29:30 MDT