I¹m also seeing similar issues on a RedHat 5.4 server,
iscsi-initiator-utils-6.2.0.871-0.10.el5.

I¹ve got a Dell Poweredge 2950 with a 2 port MLT using broadcom NICs to a
Nortel 5510 swtich. My Infortrend S16E-R1130 unit is connected using 8 Gbig
ports (it has 2 controllers) ­ it¹s all on the same switch and same private
IP subnet.

The storage is configured using the multipathd to load balance.

I¹m seeing similar TCP ³weirdness² as the other posts mention as  well as
the below errors.

Nov  2 08:15:14 backup kernel:  connection28:0: ping timeout of 5 secs
expired, recv timeout 5, last rx 4295808782, last ping 4295813782, now
4295818782
Nov  2 08:15:14 backup kernel:  connection28:0: detected conn error (1011)
Nov  2 08:15:14 backup kernel:  connection27:0: ping timeout of 5 secs
expired, recv timeout 5, last rx 4295808790, last ping 4295813790, now
4295818790
Nov  2 08:15:14 backup kernel:  connection27:0: detected conn error (1011)
Nov  2 08:15:14 backup kernel:  connection33:0: ping timeout of 5 secs
expired, recv timeout 5, last rx 4295808796, last ping 4295813796, now
4295818796
Nov  2 08:15:14 backup kernel:  connection33:0: detected conn error (1011)

The performance isn¹t what I¹d expect:

[root@backup backup]# dd if=/dev/zero of=test count=5000000
5000000+0 records in
5000000+0 records out
2560000000 bytes (2.6 GB) copied, 280.271 seconds, 9.1 MB/s

I just did a rudimentary test to see how Windows server 2008 would perform
with the same storage, and configured to access just via one interface, I
was able to sustain 20MB/s ­ so it would ³appear² to be a Linux-related
issue.

Matthew