We are seeing a number of performance problems with Netware NW65SP6 (plus
patches) servers connecting to Netapp filers via iscsi.

Looking at the LAN trace, for reads we see an ISCSI request pdu sent from
the Netware server to the Netapp - the Netapp returns data, then the next
ISCSI request is sent to the Netapp. So the disk requests are made in a
serial fashion & can result in queue build up on the Netware server.

I've been trying to get some form of "parallelism" into the requests that
the Netware initiatior makes - but without success.

I've found that the Netapp defaults to 1 TCP connection per iscsi session, I
've changed this to 4 - seen that this is what is negotiated at iscsi
session establishment - but still only 1 TCP connection is made from the
Netware initiator.

I've tried experimenting volumes/pools spread across multiple LUNs to see if
that would enable some form of parallelism - but to no avail.

I've seen parallelisnm from a Netware initiator to a Netware target - but
that is only because the Netware target advertises itself as multiple iscsi
targets to an initiator (1 per iscsi partition on the target).

I'd be grateful for any thoughts eg

Has anyone looked at the relative performances of Netware iscsi vs OES linux
iscsi for NSS volumes ? We also are about to start investigating VMWARE
(using iscsi HBAs) - might VMWARE initiator (native or HBA) be better than
Netware ?

BTW I've read about suggested options for TCP (minshall & so forth) - looked
at LAN traces - tried these - but its made no difference in our environment.

Thanks in advance for for any advice/help