MLT - a problem with connection division 1

[guygt]

Hi all,
I have a little problem with my 8600 on the company.
We have an 8600 designated as a core switch (2 units).
We have a double (2 gig) connection to each environment, which by itself consists of two Passports (8600 or 8300).
The connection is made with MLT to each environment, consists of 2 gigs to every core switch.
The problem is, that somehow, in some environmental switches, we see that one of the links is rather lowed octets (let's say just for the example 7000 octets) and his colleague is much higher (let's say again 17000000 octets),
We have checked the SW version and the HW revision of the cards - we don't see anything special.
We have 3.2.2 on the core switches.
We have 3.2.2 -3.7.5 on the environmental switches.
Does anyone know why does this phenomenal happen?
Did you ever run into this problem?
Does anyone have an Idea?


10X
G.


You have to dream of being a Pro before you become one !

 

[lambent]

Are you actually using SMLT?
And do you mean that your issue is a load-balancing issue within the MLT group?

 

[guygt]

Hi lambent,
I have configuration going like this:
1. MLT only between the environmental Switches andf the Core Switches.
2. This MLT Carries ospf redundancy.
3. I mean that the issue is indeed Load-Balancing between the MLT links.

You have to dream of being a Pro before you become one !

 

[InDenial]

expecting both links to carry the same amount (or almost the same) is a wrong expectation.

wich route it takes depends on source ip or mac-address I believe.

We had the same problem where we made a 4 Gig MLT and we noticed that all data went over just 1 link. Nortel has this algoritm which you can use to see over wich link the data goes.

check this: Link

InDenial

 

[guygt]

10X,
InDenial - I have checked the link, and I think you are right.


You have to dream of being a Pro before you become one !

 

[dane775]

Sounds like you found an answer, but I'd just like to add that 3.2.2.2 code has a ton of issues with the PP8600 (not nearly as many problems for 8100s).

There are a bunch of unpublished CRs that relate to 3222. I wouldn't even think about SMLT with PP8600s running that code. Nortel had multiple engineers camped out at one of our big customer sites for weeks dealing with 3222 code issues.

I'd get off that ASAP.

 

[MrTuo]

3.2.2.x code isn't really supported. You should opt to go up to 8691cpu or 8692's and at least the 3.5 code stream. On that 3.2.2.x stuff you're going to experience all sorts of anomalys and that version of the code has memory leaks too as in ramps up but the ram utilization never expunges....