Bay Stack 5510 Communication Problem from 100 Mbit/s to 1000 Mbit/s

[xyCruiseryx]

hi,
i have a little problem with a Bay 5510.
A connection from a 100 Mbit/s Client to 1000 Mbit/s Server is very slow.
The connection duration from 100 MBit/s to 1000 Mbit/s is higher as 100Mbit/s to 100 Mbit/s.

But when the client works with 1000 Mbit/s and the Server works with 100 Mbit/s the connection is ok.

The Connections are very surprising.

Client Server

1000 Mbit NIC to 1000 Mbit NIC = OK
1000 Mbit NIC to 100 Mbit(if configure) NIC = OK
100 Mbit NIC to 100 Mbit NIC (if configure) = OK
100 Mbit NIC to 1000 Mbit NIC = Fail

FW: 1.0.0.19/Software 5510: 3.0.0.0 Update 4.0.0.19 Result: no effect

NIC: Intel pro 1000 mit SW: 8.1.8.0 10/100/1000
Intel pro CT mit SW: 7.2.1.7 10/100/1000
3 Com 3c905CTX 10/100 mit SW: 3.21 Update auf SW: 4.31 (no effect) 10/100


Have anyone a idea waht the problem is ???

 

[anthonyanderberg]

Its worth checking that the switch and NIC agree on the duplex of the link when they're set to auto, your description sounds suspiciously like one side is talking at full duplex while the other is talking half duplex.

 

[xyCruiseryx]

No the Full an Half Modes are the same. by 100 and 1000 Mbit/s


Anyone a idea too ????

 

[oak2278]

If you want 1000->100 or 1000->10 to work fine with large TCP window sizes, you will need to make a couple of adjustments.

After updating to 4.0, and doing a default on the switches with the new code, do the following.

-Go in to the command line interface
-Type the following:
config t
qos agent buffer max
qos agent queue-set 4
end
logout
-Reboot the switch(s)

That will completely resolve these issues.

Another thing to think about is setting spanning tree to fast start if you use it.

 

[barasch]

We had this problem. Have you sniffed the network. We had huge number of retransmissions.

Solution was upgrading the software on the 5510 to 4.1 and resetting to factory defaults. Then add any vlan or other stuff.

 

[Jynxx]

The reason that loading post-4.0 code on the switch and factory defaulting it worked, is that they changed the default queue number to 2, vs 8...so the default queue packet buffer is larger, thus working.

Yes, you WILL see retransmits. When running pre-4.0 (or post 4.0 with queue set 5-8), the buffers will fill and the switch will drop packets, causing retransmits, then, obviously, more slowness.

If you use QoS within your network, use queue set 4, which most closely mimics all other L2 Nortel switches in regards to queues and buffer memory.

I have tested all queue sets, with buffer size set to max, and queues 1-4 work just fine. 5-6 are iffy, 7-8 are unusable...

Any more questions on this, just ask. We had, at one time 11 CR's open with Nortel on the 5500 product line...had the global product manager for the 5500's involved, etc, etc...it was ugly for a while. But regardless, I know these dumb things inside and out because of that...whether it is a blessing or a curse.