RE: load balancing

There are a number of possible ways to handle either load distribution (i.e.
DNS round robin) or true load balancing (i.e. Cisco Local Director or a L4
switch), each with their own benefits and problems. What your purpose
behind this will also affect which option you look at.

For example, if you're simply attempting to spread the load across two
servers and you don't have a number of downstream caches, then DNS round
robin might do it for you. It is inexpensive in terms of real dollar cost
and relatively simple from a design point of view, but it doesn't give real
failover capability, which means that if one of the servers is down, then
approximately 50% of your clients will get errors (assuming a two-cache
system). Also, since it is based on DNS and not on anything that has
visibility to the performance of the caches, it can't do load balancing,
only distribution.

If all you need is failover, you can use the JavaScript AutoConfig scripts.
You can point your browsers to use the PAC file instead of manually
configuring proxy servers. Inside the PAC, you can specify both of your
proxies and the browser will try to use the first one all the time and use
the second one only if the first fails. Like the DNS round robin, it is
basically free. Unlike round robin, it only provides failover, it doesn't
do any load distribution at all.

Finally, if you are looking for something that provides a bit of both and
are willing to pay some money, you could get a level-4 switch or a Cisco
Local Director. These devices work at the network layer and can be
configured with rules as to how to distribute traffic. I only have
experience with the LD, so I'll have to address from that side, but there
are similarities between the two types. On the LD, you can set it up so
that you're doing true load balancing (i.e. distribute new sessions to the
machine with the fastest response time or the fewest established
connections, or specify percent rules for normal operation and for failure
states, etc.) and it can also handle failure states automatically switching
all traffic to the functioning box within a matter of seconds. Of course,
the downsides here are high dollar cost and increased complexity. There are
wierd technical problems that can occur doing this, so I don't recommend it
unless you really know what you're doing, both on the caching side and on
the network side.

Hope this helps,

Jon

> -----Original Message-----
> From: Burak Demircan [SMTP:burak.demircan@mbturk.mercedes-benz.com]
> Sent: Thursday, September 16, 1999 7:02 AM
> To: squid-users@ircache.net
> Subject: load balancing
>
> How can i do load balancing on 2 squid
> proxy servers? Or is it possible?
> thank you
>
Received on Thu Sep 16 1999 - 08:36:54 MDT