VNX5300 long iSCSI path failover times on vSphere 5

hi all,

I'm trying to find the reason for long path failover times in a vSphere 5 environment. The setups looks as following:

VNX 5300 with 2x 10G iSCSI Ports on the SPs

     A0 and B0 are in the same IP Range and the same VLAN - 9000 MTU configured

     A1 and B1 are in a different IP Range and a different VLAN - 9000 MTU configured

     latest Firmware on the SPs (05.31.000.5.704)

     all ESXi Hosts registered as Clariion Open with failover mode 4 - 4 paths per registered ESXi Host

Nexus 5548UP Switch inbetween the VNX and the HP Servers running vSphere 5 ESXi Hosts - jumbo Frames enabled

ESXi Hosts with SW iSCSI initiators and 10G NICs (no PowerpathVE)

     two VMK Interfaces each in the VLAN of two iSCSI Ports and an IP address in the corresponding IP Range of the 2 iSCSI Ports on the VNX

     VMW_SATP_ALUA_CX is set as the NMP

     each VMK is on its own vSwitch with the 9000 MTU settings on the VMK and the vSwitch itself

     iSCSI VMK to VMNIC mapping done under the SW- iSCSI initator in ESX and each VMK has only one active VMNIC (others are unused)

     delayed ACK option unchecked in the advanced options for iSCSI

the ESX Hosts can see the VNX, all masked LUNs, VAAI funtionalities, test VM running and so on - so far everything was perfekt.

the problem is when we were testing the path failover times by shuting the correspoding iSCSI interfaces (one at a time - either on the ESX Host side or on the VNX side) on the Nexus we experienced long periods (20 - 30 sec)  where the datastores on the Hosts were not accessible (did a simple ls on the ESXi host to verify) 

the load on the box was non-existent (running 1 test-VM on the whole VNX and the ESXi Host)

manually switching the preferred path (when using the fixed policy in VMware) worked just fine.

the failover times are the same for fixed and RoundRobin policies and also when shuting the host interfaces or storage interfaces on the Nexus.

my questions regarding this issue are:

* what are the expected path failover times on iSCSI block accessed LUNs - are there any documents describing those?

* good solution(s)/hints for decreasing the failover times would be amazing

thank you in advance for any help

Best Regards,

Lukas