[olug] Border management / server farms

[Jay Hannah]

More info on my evil plot:
- Web tier: Linux/Apache/Perl applications
- DB tier: Linux/mySQL replication?
- Out front: a load balancer of some kind (thanks for all the 
suggestions!)

On Wednesday, February 4, 2004, at 10:14 AM, Nick Walter wrote:
> Oddly enough, I was in the UK setting up something like this last week.

Why didn't you invite me so I could learn? -grin-

> The load balancing among tomcat would keep requests going to live
> servers in the event of an application server outage,

Do X servers in Tomcat share the same IP address and then poll each 
other to see which peers are alive? ... I don't understand how this 
works without front-end servers if one of the servers locks up at the 
application layer (still pingable, but Tomcat not successfully 
processing?).

I need to go research Tomcat load balancing...

On Wednesday, February 4, 2004, at 10:22 AM, Rod Hurley wrote:
> Another plan would be to go with a VMWare product.  In this instance
> you would virtually divide your web into 2,3,4 or more O/S 's with the
> same apps, or different ones running at the same time on the same box.
> Needs plenty of RAM and I recommend dual (or better) processors, but
> much more cost efficient than buying 4 servers (compared to one server
> with 4 virtual servers.)

I'm worried about the cost of scaling expensive hardware and am 
thinking on the web tier I'll be better off running 4+ truly craptastic 
machines behind some "border" machine(s) which does the load balancing; 
rather than expensive (highly reliable) hardware which is too expensive 
to scale...

On the database tier I'm wondering if the same thing applies, wondering 
if I can get away w/ a bunch of cheap Intel boxes running mySQL 
replication, with a device in front routing traffic to the node(s) 
currently known to be online.

Citrix Metaframe? I thought that was for Win* environments?

On Wednesday, February 4, 2004, at 10:39 AM, Jay Swackhamer wrote:
> There is some information that you might find usefull at
> http://www.linuxvirtualserver.org/
> http://lcic.org/load_balancing.html

Ooo. Thanks. Must read.

On Wednesday, February 4, 2004, at 10:39 AM, Jay Swackhamer wrote:
> A simple solution would be to have the name server do a round-robin
> assignment of names so that the IP addresses of active servers get 
> cycled
> through, and have the name server actively check the servers and 
> re-move
> the non-responding servers from the rotation.

"actively check"? How would I do that? Roll my own?

I'd still get screwed when servers cache my DNS entries, though. 6 
hours of TTL on some name servers would be bad.

On Wednesday, February 4, 2004, at 12:53 PM, Christopher Cashell wrote:
> Are you looking for a generic solution for use with an existing
> application?  Or do you have the ability to customize/develop the
> application with high-availability in mind?

The latter.

> If the application is not yet finished, or not yet started, you might
> want to look into using an "application server" as the framework for
> your project.  Many of them natively support database pooling, and 
> could
> greatly simplify things for you.

Hmm... Are there application server options for my Linux/Apache/Perl 
applications?

Another hot tip I got for Perl/Apache/mysql load balancing off list:
    http://www.ultramonkey.org/

Let the research and fiddling begin...

j