[Zope] Advice needed: load balancing wih ZEO and Apache on So laris.

[Tony McDonald]

On 25/8/01 8:55 pm, "sean.upton@uniontrib.com" <sean.upton@uniontrib.com>
wrote:

> Sounds like you need something to distribute and/or balance the load; 3
> suggestions:
> 
> 1 - HTTP Proxy Load Distribution - Squid does this, round-robin, put squid
> in front of your ZEO farm (this is what I do).  I don't think that mod_proxy
> is capable of doing this.

Sean, Phil,
Many thanks for the help. I couldn't reply earlier due to family matters.

I downloaded squid and spent quite some time with it. Unfortunately, it
seems way too complex for me. However, as it has the http-accelerator and
other facilities, I'll probably look at it again once I get some free time.

> 2 - Layer 4 switch (Cisco LocalDirector, Intel Netstructure, etc) - not
> cheap, but offers some features like a choice between true balancing, outage
> detection, and funky routing techniques like out-of-path return for quicker
> network performance.

At the moment we're strapped for cash, so I can't use this option - although
as we're going to be putting some firewalls in front of our boxes, I *may*
be able to do this in the future.

> 3 - LVS - Linux Virtual Server Project, attempts to do in software what # 2
> does.
> 

Had a look at this, but we're a Solaris shop and there seemed too many
kernel patches needed to do this.

> Personally, I'm in favor of squid because it is cheap and easy, if all your
> machines are of equal weight in performance terms.  My company uses squid in
> front of multiple ZEO clients, bypassing Apache altogether for access to
> Zope; we use pyredir as a redirector to rewrite URLs like you would with
> mod_rewrite; Squid is very nice as a Zope virtual host front-end, and
> provides ACLs for security purposes, like blocking out the ZMI from public
> access.  

This is the main reason I spent that time looking at squid - the security
aspect. But, we've built up quite a bit of experience with Apaches
ReWriteRules and have a requirement for PHP and (cough) Perl scripts to run
alongside our Zope sites as well.

> Getting load balancing working in squid is simply a matter of
> compiling it with a flag to specify you want external DNS resolution support
> and using squid's built-in name resolving program dnsserver, which will look
> at /etc/hosts and round-robin among multiple IPs that have the same
> hostname.  One caveat: if you put Squid (or any high-volume TCP app) on
> Solaris, make sure to tune your TCP connection buffers, so Solaris doesn't
> choke; this is one (albeit trivial) reason we use Linux 2.4 for our proxies
> instead of Solaris.
> 

Ah - thanks for that - I'll do the /etc/system thing with the TCP/IP
parameters.

> Since squid round-robin's it doesn't deal with node reliability problems,
> which means you will have to rely upon another HA mechanism on the ZEO
> client nodes themselves, like heartbeat from the Linux-HA project, which at
> some point will ported to Solaris, from what I hear.
> 

We're going to be getting some high availability software that does
heartbeat monitoring so I think I'm ok there (but see *** below).

> If you have a decent budget, but not a lot of time, look at a L4 switch (we
> use for another application, and like, the Intel Netstrcuture 7140);
> otherwise, if you have budget constraints consider LVS or Squid, with Squid
> being the likely easiest path in terms of setup time.
> 
> Sean
> 

Thanks very much for the info Sean. Although I'll probably not be using
squid at the moment - all this is very helpful. In the end I used the method
that Phil mentioned;

> Anyway on with the show:
> 
> Take a look at this document,
> http://httpd.apache.org/docs-2.0/mod/mod_rewrite.html.  Search for
> 'Randomised Plain Text' and there's your recipe.
> 
> Basically, you need to create a file that has these lines in, call it
> map.txt (the name doesn't really matter though):
> 
> localhost    port1|port2|port3
> 
> replace port1 .. port 3 with the exact ports you are using:
> 
> localhost    8080|8081|8082
> 
> You can put as many options on this line as you need.
> 
> Then change your final rewrite rule to be like this:
> 
>   RewriteMap servers rnd:/path/to/file/map.txt
>   RewriteRule ^/(.*)
> http://localhost:$(servers:localhost)/VirtualHostBase/http/myserver.ncl.ac.u
> k:80/VirtualHostRoot/$1 [P]
> 
> This will put a random port number into the line thereby giving you pseudo
> 'round-robin' functionality.  The ${servers:localhost} is the clever part,
> 'servers' is the name of the map, and localhost is a parameter telling the
> map which line of the map.txt file to choose from.

as I have quite some experience with Apache and Rewriterules.

*** We're only going to have two machines in our 'cluster', but they are
multi processor machines. The reason I'm having to go through all these
hoops is basically down to the poor performance of Sparc chips. My pystone
ratings on our new server are about 4500, whilst my own TiPB is 6500 and a
PIII 700MHz is about 10,500. Therefore I'm trying to squeeze more
performance out of our boxes.

Once again guys, thanks for the help.

Now all I need to do is figure out how to do Core Session Tracking with ZEO
(I know there are HowTos - but they're not that transparent to my poor head!
:)

Cheers,
Tone.
-- 
Dr Tony McDonald,  Assistant Director, FMCC, http://www.fmcc.org.uk/
The Medical School, Newcastle University Tel: +44 191 243 6140
A Zope list for UK HE/FE  http://www.fmcc.org.uk/mailman/listinfo/zope