iSCSI on RHAS4 (or CentOS4)

I'm wondering what other experiences people have have using the iSCSI
initiator RPM which comes with RHAS4 (or CentOS4). I'm having difficulty,
but I don't know if it is an issue with the software on the Linux side
or the SAN.

All my testing described here is on a single Linux box as an initiator.
There's a direct GigE connect to the iSCSI target (the SAN); no switch
involved. I allocated a /30 out of private address space for this.

I've tested the connection with utilities like flood pings, and all is
well.

When I've a single reader or writer process, all is well. When I've
multiple processes writing to the same volume mounted from iSCSI, all
is well.

But when I've multiple processes *reading*, performance becomes
incredibly poor and /var/log/messages tells me:

Mar 10 11:24:10 168 kernel: iscsi-sfnet:host1: Received opcode 0, ahs length 1, itt 67132672. Dropping, additional header segments not supported by this driver version.
Mar 10 11:24:10 168 kernel: iscsi-sfnet:host1: Session dropped
Mar 10 11:24:10 168 kernel: iscsi-sfnet:host1: Failing command cdb 0x28 task 109614 with return code = 0x20000
Mar 10 11:24:10 168 kernel: iscsi-sfnet:host1: Failing command cdb 0x28 task 109615 with return code = 0x20000
Mar 10 11:24:10 168 kernel: iscsi-sfnet:host1: Waiting 2 seconds before next login attempt
Mar 10 11:24:13 168 kernel: iscsi-sfnet:host1: Session established

many times (sometimes w/o the "Received opcode" line).

The RPMs in question are:

iscsi-initiator-utils-4.0.3.0-3
kernel-smp-2.6.9-34.EL

Multiple readers can be achieved as easily as:

dd if=/mnt/raid10/001 of=/mnt/local0/001.copy.1 &
dd if=/mnt/raid10/001 of=/mnt/local0/001.copy.2 &

Are others having difficulties like this?

- Andrew