Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[paul]

But it seems i can't start a new topic? Do i miss something?

Anyway,

First, a small description:

X-Stack switch1 = 2 x DXS-3326GSR + 6 x DGS-3324SR
X-Stack switch2 = 2 x DXS-3350SR

Both X-stacks are in ring configuration.
Both X-stacks have firmware 4.40-B04

I made a LACP link between the two X-Stacks, using four ports on both sides, on different stack units.

When i connect the X-Stacks with one (fiber) uplink, everything works fine. But link is limited to 1GB.
When i connect the X-Stacks with two, three or four uplinks, communication is still OK. But, the CPU load of the first X-Stack goes rocketing skywards, almost 95%... that's not cool, since management comes to a hold (and so does answering to SNMP calls). And that triggers my nagios monitoring... making me very nervous, this switch is the core of our network, there's 900 employees depending on it (yes, we're using citrix, so no network - nothing goes. Exept for my workstation, that's a linux system :-)..

I already asked D-Link tech support about this matter, but i got no response at all...

anybody an idea?


regards,

Paul

[Fatman]

Do you see this problem if you use all LACP members from one stack switch on each side?

[paul]

I created a LACP group with four ports on both 'sides'
To create as much redundany as possible, i created this group with two ports on unit 1 en two ports on unit 2 of the X-Stack. I did this on both sides.
So the LACP group on X-Stack 1 is UNIT1-port5&6 UNIT2-port 4&5
On the other side, the LACP group is UNIT1 port2&3, UNIT2 port 1&2
Looks a bit like this:

X-Stack 1...........................................X-Stack 2
UNIT1-Port 4....................................................................Unit1-port 1
UNIT1-Port 5-LACP-Masterport <----->LACP-Masterport-UNIT1-port 2
UNIT1-Port 6-LACP port.......... <----->...........LACP-port-UNIT1 port 3
UNIT1-Port 7....................................................................Unit1-port 4
etc..
UNIT2-Port 4-LACP port.......... <----->...........LACP-port-UNIT2 port 1
UNIT2-Port 5-LACP port.......... <----->...........LACP-port-UNIT2 port 2
UNIT2-Port 6...................................................................Unit2-port 4
etc...

The high CPU load occurred when i connected more than one port of the LACP group.
There was no networkloop (everything kept working) but the CPU went rocketing...
And with such a high CPU load SNMP stops working...
After disconnecting the 'extra' links the CPU load went normal..

BTW, all the links between stack 1 and 2 are fibre, as it is the link between two server-rooms...

I'll have to get some spare-switches to do some further testing, as the link is now (with only 1GB) operational... so i'll have to be very sure changes work instantly and correct before making changes to these production-switches.


with kind regards,

Paul

[Fatman]

Do you see this problem if you use all LACP members from one stack switch on each side?

[paul]

I'm not quite sure what you mean...

I see the problem occur when i actually use more than 1 port between the X-Stack switches.
Creating the LACP trunk is no problem. Using it is...
So, if i use one port from the LACP trunk, no problems
If i connect the second cable the CPU-load goes whammmm (and the third and fourth do not make a difference, the CPU load is already rocketing.

So i guess the answer on your question should be no, i also see this problem when i'm not using all the members.

regards,

Paul

[Fatman]

Ok, I will clarify then.

If this is your stack

Stack-1-Switch-1
Stack-1-Switch-2
Stack-1-Switch-3
Stack-1-Switch-4

Do you see an issue if your LACP configurations and connections are as follows.
Stack-1-Switch-1 Port 1 <-> Stack-2-Switch-1 Port 1
Stack-1-Switch-1 Port 2 <-> Stack-2-Switch-1 Port 2
Stack-1-Switch-1 Port 3 <-> Stack-2-Switch-1 Port 3
Stack-1-Switch-1 Port 4 <-> Stack-2-Switch-1 Port 4

I realize this is not what you are trying to accomplish, but I need to know if LACP traffic across the actual stack connections are what is causing this behaviour.

If this does not make any sense, or if this line of questioning is going nowhere please call BCS Support at 1 877 354 6555.

We don't even have a board for this product, largely to avoid fielding large scale problems like this in an environment where communication is vastly asynchronous and easy concepts can be difficult to communicate.