[dev] Re: [cvs] commit: horde/docs CHANGES horde/lib Horde.php

[Atif Ghaffar]

Rich Lafferty wrote:

> That doesn't matter.  The proper way to load-balance dynamic content
> is with a load balancer that understands that it has to make up for
> HTTP's statelessness. If yours doesn't, it is broken for what you
> appear to want to use it for, and fixing the effects of its brokenness
> elsewhere does not stop it from being broken. Ensuring that no Web
> applications ever use files is *not* the way to work around a load
> balancer that is not designed to handle dynamic content.
> ("Content-aware" is the marketing term, IIRC.)
Rich, here you are only talking about HTTP loadbalancers :). 


I use the term load balncer for anything that can distribute requests to
multiple machines.

Example ipmasqadm which forwards port 80 to different machines or even a
round robin setup.

> Relying on the filesystem is a perfectly legitimate assumption, since
> that is precisely the problem that load balancers are supposed to
> solve. (Otherwise you wouldn't need a load balancer; round-robin DNS
> would accomplish the same thing.)

We setup once a webapp cluster where there was almost no rw filesystem
mounted.
In that case attempting to write on a filesystem (other than asking the
filesystem to give tmpfilehandle) would have had been problamatic.


With most SAN based clusters, it is rare to see multiple machines having
rw access to filesystem. same is true when the file system is being
imported over NFS or other Network file systems.


Anyway, I didnt mean to critisize on your practices.