Is iSCSI SAN still a spof in a cluster? Can it be highly available

Consider whether you really need hot failover. If you're worrying about price at the level of an entry level SAN then consider whether you really have a business case for that recovery model. How expensive is your downtime for an outage?

If the cost of an outage justifies the cost of a decent SAN, buy it and don't penny pinch. Otherwise, consider other failover models. If your downtime is not so valuable, you can probably tolerate a hot standby model where the database is replicated to another server with local disk. This takes longer to recover but does not need shared disk storage. If this works for you then you don't need a SAN and the local disk on the servers will probably be much cheaper.

Another option would be the secondhand market. You can get a re-certified second hand Clariion CX200 or CX300 (which would probably do what you're after) for just a few thousand dollars. Re-certified hardware qualifies for vendor support and can be purchased through various outfits such as www.berkcom.com or www.bltserv.com.

(Disclaimer: I have no affiliation with either vendor but am a satisfied customer of BLT Services. Berkcom was recommended to me when I needed something that BLT didn't have).

Most of the major points are covered in the answers to the post gbjbaanb referenced in his comment but I would add some more to specifically address your questions.

Adding any PS series array to this will improve your performance - provided you let the EQL automatation to do its thing. A PS6000XV configured in RAID 10 will help most clearly but a 10K SAS PS6000X or even a 7.2k SATA II PS6000E will improve the overall IOPS available, add network paths and bandwidth. Any one of them will add at least 1000 IOPs to the overall capacity of your environment.

My opinion on Equallogic configurations is that you should avoid dedicating arrays to a single purpose - you will almost always find that overall performance is better if you allow the Equallogic Group to distribute volumes over as many arrays as it can, that way your systems can scale out a lot better. By dedicating arrays to a single volume or function you are limiting the total number of paths, bandwidth and amountof controller CPU available for those volumes.

The average IOPS (178) indicates to me that you're not really pushing the system's limits at all. Even with 7200rpm drives, only one controller (which I hope you don't have) a single array and RAID 5 you should not be stressing even a single PS array at that level, although I may be missing something here. You appear to have a 70:30 Read write ratio so RAID 50 is probably optimal for this use case.

I'm intrigued by the 20ms read latency number, your writes at 2ms seem to be OK but your reads are much higher than I'd expect. Are you sure there is nothing else going on? Disk failures\replication\huge numbers of snapshots?