Unsolved
This post is more than 5 years old
8 Posts
0
28590
July 27th, 2012 16:00
Powerconnect 5548 Ring Topology issues
We have 4 Powerconnect 5548 switches stacked in a ring topology per page 46 of the User's Guide. For simplicity's sake I have attached an image and labeled each switch A-D. I appreciate you taking the time to look at this but please read it completely before commenting.
With switch A as the master (A=1,B=2,C=3,D=4) there is no traffic passing on the pink cable (connecting switch C and D) and the activity lights aren't doing anything. When I connect a laptop to switch C and have it PXE boot from a server on switch D the image load status is choppy and appears to come across in bursts. File transfer between devices connected to switch C and switch D are also slow. When I disconnect the cable connecting C and D these problems disappear completely. If I disconnect the cable from A to D the problems also disappear.
I changed the configs on the switches to reverse their stack order (without physically changing anything). So now switch D is the master (A=4,B=3,C=2,D=1) and it causes the orange cable (connecting A to B) to stop passing traffic. I tried other configs like that and every time the connection between switch 3 and switch 4 sits dormant.
The user guide says "In a ring topology all units in the stack are connected to each other, forming a circle. Each unit in the stack accepts data and sends it to the unit to which it is attached. The packet continues through the stack until it reaches its destination. The system discovers the optimal path on which to send traffic." but this is clearly not what they're doing. I could understand if the user guide was simply misleading and meant to state that the switches will find the shortest route back to the master but even if that were the case it doesn't explain why the traffic gets jacked up when all four cables are connected.
We have the latest firmware running on each switch and our config is pretty basic.
To reiterate, it seems that no matter how I stack these switches they refuse to pass traffic on the cable connecting switch #3 and switch #4.
KyleReese
8 Posts
0
July 27th, 2012 20:00
Simplified: no matter how I configure this stack, switch #3 and switch #4 won't communicate directly with each other if the stack is configured in a ring topology. This contradicts the Dell documentation stating that the system will determine the optimal path to send traffic. Furthermore, traffic between a device on switch #3 and a device on switch #4 is noticeably slower than that of any other pair of devices unless the ring topology is broken. So not only is the optimal path not being found or used, connection speed is actually being degraded.
KyleReese
8 Posts
0
July 30th, 2012 11:00
The switches are all PC 5548, they're all running SW 4.1.0.1, and the cables are all identical. We will try lowering the stack count to 3 tonight to test that.
We have tried physically swapping the cables around but regardless of the physical setup or the "running-config" setup the 3 to 4 link sits unused.
Can you verify that in a loop configuration all traffic is supposed to be routed the in the most effecient manner?
KyleReese
8 Posts
0
July 31st, 2012 04:00
In a stack of 3 the cable from SW2 to SW3 is ignored.
KyleReese
8 Posts
0
August 6th, 2012 20:00
Using Dell's HDMI cables yields the exact same behavior...the link from SW3 to SW4 has no activity (beyond the first several seconds of plugging it in when it's presumably finding the fastest path through everything). Can you provide a link to any documentation stating emphatically that the switches will find the shortest route between ports to route data?
KyleReese
8 Posts
0
August 7th, 2012 15:00
That quote from the document is in my original question. It doesn't *exactly say that the optimal path will be found from port to port. We replaced our cables with Dell cables that are doing the exact same thing (not using SW3 to SW4) regardless of how we renumber them electronically. So what gives?
OSI-Bruce
1 Message
0
November 20th, 2012 15:00
Don't know if you ever got this answered.
That is how it is supposed to work in a ring topology, whether stacking or using Spanning Tree Protocol (STP). The switches automatically select the "path of least cost" (fewest hops, etc...) as the default.
To get from sw-A to sw-D there is a lower "cost" on the direct connection than going through sw-B + sw-C to sw-D. A ring is about redundancy, if you disconnect any one of the cables, the path will be rerouted to the next least cost available path.
Examples:
Disconnect the A to D cable - The path to D will go through B and C to get there,
Disconnect the B to C cable - The path to C from A will go through D to get there.
Hope this makes it clearer.
KyleReese
8 Posts
0
November 21st, 2012 07:00
What you're describing is that any given switch will find the shortest path to the primary switch and that the shortest path from port to port isn't calculated.
If correct, that would explain the behavior we're seeing:
Traffic from SW3/p7 to SW4/p12 goes SW3 -> SW2 -> SW1 -> SW4 (ignoring the obvious SW3 -> SW4 path)
If that's the case then it stands to reason this is also true:
Traffic from SW3/p8 to SW2/p4 goes SW3 -> SW2 -> SW1 -> SW2
Are you stating emphatically that in a ring topology one cable is never going to carry traffic?