Direct attach ESX hosts to CX4-120 not support or BAD?--------need help

Hi:

We are at the ¿final decision¿ stage to buy an EMC CX4-120 box.
But something came up from the EMC (?) reseller which I really need the EMC engineer or anyone knew the CX4-120 to clarify the confusing.

I worked with SAN but not EMC SAN.


Background:
The CX4-120 is for a VMware vSphere farm with 4 ESX 4 hosts. It has 50 + VM and might grow to 100 +
All ESX hosts are HP DL380 G6.
Due to we only have 4 ESX hosts, we do not want to use FC switches but just want to direct attach those four hosts to the SAN box (two HBA per host).
I think that it is a supported configuration (Per the EMC and VMware¿s doc).

But the reseller tell me (please see the following) that it is not support and it is a BAD idea.
I think what he said regarding to the trespass, the way ESX hosts see the LUN, etc were not correct.

But I could be wrong, so if any EMC engineer or anyone knew EMC CX4-120 can help us to get a ¿official¿ clarification will be great appreciated

Thanks for the help!!!


The following is from the reseller:
------------------------------------------------------------------------------------------------------------------------
But aside from the port limitation, the greatest reason to NOT directly attach servers to the storage processor is high-availability. In the event of a path failure (bad cable or bad HBA in a server), the failed link directly to the storage processor would cause a LUN trespass to the other storage processor. In a physical server environment this isn't a huge issue since servers don't share LUNs with other servers (except in a clustered environment). However, Vmware shares all it's LUNs across all servers in the Vmware farm, so a single path failure for one server would cause the LUN to move needlessly to the other storage processor. It could then cause a "ping-pong" effect, also known as "path trashing", where the other VMware servers don't see a problem with the path, and they in turn try to move the LUN back to the original storage processor. Then the server with the failed path tries to move it back (etc etc etc). There are ways to avoid the path thrashing but no way to avoid the initial (and unnecessary) LUN trespass without a different SAN fabric architecture.



Alternatively, with an FC switch in the middle (between servers and SAN), each switch would have connectivity to BOTH storage processors, so in the event of a single HBA or cable failure, there would still be a path to each storage processor. So a LUN trespass would only occur if a storage processor truly died or if BOTH paths to the specific storage processor went dead.



This is somewhat difficult to fully describe without a picture...