MLT problem on 5510 and 450

[m1227]

Hi all,

Here is the story and make it simple.

I had a Nortel networks with couples of 450-24Ts (Named as A450 and B450) and 5510-48T (Named as A5510) in single VLAN1.
Connection as below:
A5510 ==(2 x 100M MLT)==> A450 ==(2 x 100M MLT) ==> B450
Workstation 1 (WS1) connected to A5510.
Server 1 (SVR1) and Server 2 (SVR2) connected to A450.
Workstation 2 (WS2) and Workstation 3 (WS3) connected to B450.

Normal stage, all workstations and servers can talk/ping with each others.

On problem stage;
WS2 cannot talk to SVR1 and SVR2 but WS3 can reach SVR1 and SVR2. (WS2 and WS3 in same B450 switch)
At the same time, WS1 can talk to WS2 and vice vise without problem.

After many times of trying, I found a temporary solution.
Physical disconnect one leg of MLT between A5510 and A450, wait 1 minutes and pull back.
About 10 to 20 hours, another workstation (WS4) at B450 have the same problem again.

So, I think some kind of Spanning-Tree problem happened on the networks.

Are there any expertise over here can give me a hint to trouble shoot this problem.

Thanks in advance.

 

[m1227]

More information:
BS5510 using SW:v4.2.1.004
BS450 using SW:v4.4.0.6

 

[anthonyanderberg]

I've seem problems where a port will corrupt or perhaps silently drop traffic without taking the port down. Mine happened on an 8600, but the key is MLT masking the problem. Traffic across the MLT links is divided up based on last 6 bits of the MAC or IP source and destination addresses. That's why some traffic works while other traffic fails, one of the ports or links is bad and only the traffic across it is affected. Its weird that unplugging the 5510-A450 link fixes it though - are you sure the MLT groups on the A450 are really 2 separate MLT groups? Also - do any of the port counters show any errors?

In any case you can either lookup the MLT load sharing algorithm online and do the binary math to figure out which link is bad or you can try removing one link for a while - if the problem comes back and you've pulled the good link all of B450 should be down. The nice thing is that you've got a 50/50 chance - mine was an MLT with 8 links so I just shifted which 1/8th of my users were calling until I finally found the right link.

 

[WillyHe]

Are there errors shown on the MLT ports?

Remarque's:
A 55X0 uses the source/destination algorithm to decide which link of an MLT to choose for a communication pair.
On a 4X0 or 3X0 series switch its different, the first active port will take the first local link of an MLT, the second active port will take the second local link of an MLT and so on, and then active port X will take the first local link of an MLT, ....

This means e.g. an MLT with two links, the third active PCs traffic flows over the first link of the MLT and the fourth PCs traffic flows over the second link of the MLT, now when the third PC is switched off the fourth active PC becomes the third and now his traffic flows over the first link of the MLT, when the first link of the MLT has connectivity problems .... the new third PC has communication problems for packets traveling this link.