clustering with a shared disk

[kc_kazan]

i was arguing with some fellows if the following is a good practice for a cluster :
1. 2 application servers accessing a shared disk.
2. load balancing for the 2 servers done through DNS ( not so clear how !).

but the main point is #1, in my opinion - i'm not a cluster'aholic , i know very little - i think that the bottleneck this way is moving from the NIC and server resources to the disk which is even worse as it acts as a single dangerous point of failure.

I'm waiting to hear some other opinions on this matter.

[likwid]

Normally, this is done with high-speed dedicated links- like fiber, that isn't used for anything but connection to the SAN. A disk inside a server is a single dangerous point of failure- having a SAN box normally comes with a bunch of network interfaces, and all sorts of redundancy and fault tolerance. This is one of the most common setups in clustering.