[plug] Apache 1.3 x 2 on the one box

[Tim Bowden]

On Mon, 2005-11-07 at 14:45 +0800, Craig Foster wrote:
> ?
>  
> 
> ________________________________
> 
> From: plug-bounces at plug.org.au on behalf of Tim Bowden
> Sent: Mon 07-Nov-05 2:04 PM
> To: plug at plug.org.au
> Subject: Re: [plug] Apache 1.3 x 2 on the one box
> 

No idea how this got to plug-bounces ????
> 
> 
> On Sat, 2005-11-05 at 17:16 +0800, Timothy White wrote:
> > I'm not 100% sure what 'rhel 3' is, but in general you can run 2
> > instances of apache on any box. You need to specify a different port
> > for one though (so one on port 80, and maybe one on 81?)
> > I think you'll find, depending on which apache 'engine' it is running,
> > it'll probably add a little bit of memory, but not CPU.
> >
> 
> We are stuck with Apache 1.3 which ships with Red Hat Enterprise Linux
> V3. Increased ram requirements I can handle.  Increased cpu is not good-
> even though I have 2 x dual cpu (xeon) dell servers.  When the cgi app
> (mapserver) gets to 100% cpu it produces incorrect results without
> throwing error reports.  Yuck.  We need it to go as far as possible
> before we get to that point.  I'm trying to work out some quick and
> dirty simple load balancing between the two machines.  
> 
> The complicating
> factor is that all traffic comes in through one of two VPN's and must go
> out the same VPN no matter which machine it gets processed on.
> 
> Thanks for the info Tim.
> 
> Tim
> 
> ----------------------------------
> 
> Have you had a look at iptables MARKing?
> 
> I've had dual homed (Dual ADSL) router setups before where traffic is given a MARK x depending on which interface it came through, and it (iptables) set to reply on the same interface... 
> 
> I'd also seriously look at redoing the CGI scripts if they're that intensive. Dual Xeon (or opteron - to save a flame war) servers should be able to cope with almost everything...
> 
> CraigF
> 
>  

no, I haven't looked at iptables solutions.  Thanks for pointing that
out.  The boxes aren't dual homed though.  I can see in my haste I
wasn't too clear- There are two boxes at the central site (think
redundancy, a recurring theme on this project), each with one VPN going
to two boxes at the other end (remote site- ultimately >10 remote
sites), with VPN connections mapping to the corresponding central boxes.
The idea is that each VPN should be on a different network for
redundancy.  (Can you see the dollars adding up?) A request can come
from either central box 1 or 2 to the corresponding box at the other
end.  At the remote site, we need to do some form of crude load
balancing between the boxes.  The system architect did have a seperate
linux box in the system to handle the VPN's and load ballancing but that
has been scrapped by another project that has changed our clients
network infrastructure (don't ask, the politics are /dreadful/).

The cgi app is mapserver which handles gis (Geographic Information
Systems) data.  Mostly the boxes will cope fine.  In testing so far we
have been able to put a load on it well in excess of what we will have
in production, however with gis analysis it is reasonably possible that
we could get several simultaneous requests that can max out the system
because of the very large data sets that have to be processed (think 5m
contour data for entire perth metro area along with high res aerial
photos and all road centrelines together with several other very large
data sets thrown in).

Does that all make sense?

Tim Bowden