network raid possible?

[Garry Turkington]

Hi,

If you're basically building a high-volume web farm then don't reinvent
the wheel and use standard deployment architectures.  Having the content
on a NFS server behind the web servers is an attractive and potentially
low cost route, even gigabit networking is relatively affordable these
days.  If fully shared storage is *truly* what you need then NFS is the
easiest option.

Also, think about the likely data access patterns.  Do you have large
amounts of static content?  If so then this could be local on each server,
leaving only the truly dynamic content requiring the shared storage.  That
likely allows you to get away with a less beefy NFS server and back-end
network.

I'd also ask you to take a very long hard look at the stated requirement
for instantaneous synchronisation across the multiple servers.  Building a
highly available web service is exceptionally difficult if the requirement
list includes phrases such as instantaneous, "fully transparent" and "no
single points of failure".  What you really need to understand is the
implication of the possible failure scenarios and just what impact that
will have on the service clients.  Often when cost and complexity are
taken into account the requirements soften somewhat.  Otherwise you end up
designing systems where you want to have users in pairs in case one
spontaneously combusts.  You can push points of failure further up and
downstream but at some point you have to accept some risk.

Regards,
Garry

Garry Turkington
garry.turkington at acm.org