When you buy a shared strorage sdevice like a SAN you are buying a system that is a 5 nines system (99.999% uptime). I'll use EMC as an example as that's what we have at the office. Each unit comes with two heads. Each Head is connected to each drive shelf with 2 cables (4 cables going to each shelf). Our fibre fabric has two director class switches. Each switch is connected to each head via 2 fibre cables. Each server is connected to each vibre switch. This creates a totally redundant system. The host OS for the EMC (which is shared by both heads) is sitting on a dedicated 5 disk RAID 5 array (and has access to the hot spare in case of a drive failure). Pretty much the only way that both heads on the EMC can go down would be a multi-drive failure of the host OS drives, or a power failure of both redundant power circuits.
If I was designing this solution for my company or a client, I would recommend the following solution. (I'll use HP Servers as those are my normal server platform. You can convert this to what ever hardware platform you normally use.)
Web Servers
2 - HP DL360 Servers. Two single core CPUs, 1 Gig of RAM. Load Ballanced for redundancy. Windows 2003 Web Edition
Storage
EMC Clarian Storage Array. 1 Shelf of 146GB 10k drives (15 drives total; 10 usable, 5 reserved for EMC OS).
SQL Cluster
2 - HP DL 380s. Two Dual Core CPUs, 4 Gigs of RAM. Windows 2003 Enterprise Edition, clustered. SQL Server 2005 Enterprise Edition (Standard Edition can be used if the features of Enterprise Edition are not needed, as the cost savings is quite large).
Exchange Cluster
2 - HP DL 380s. Two Dual Core CPUs, 4 Gigs of RAM. Windows 2003 Enterprise Edition, clustered. Exchange 2003 Enterprise (I'm not sure if Standard Edition can be clustered or not).
If they wanted to save a little money on the high end hardware I'd recommend a single three node cluster of HP DL380s so they whould be sharing a passive node as the odds of both active nodes failing at the same time are very slim.
I wouldn't recommend host based failover software such as neverfail to a client unless they were trying to use it for failover to another site, and EMC to EMC replication wasn't available.
Denny
MCSA (2003) / MCDBA (SQL 2000) / MCTS (SQL 2005) / MCITP Database Administrator (SQL 2005)
--Anything is possible. All it takes is a little research. (Me)