Packets drops and spanning tree problem

[P.M.S.Prakash]

I am observing packet and connection drops erratically in a newly setup network using procurve switches. The attached figure explains my network.

The majority of the switches are 5308 and some 2510 switches are used. Currently, 5308 is acting as a L3 switch, floor distribution switch and access switches. In some places, 7x2510 switches are cascaded serially, instead of 5308 switches.

The VLANs are going to be spread across all the floors. I am using the primary VLAN as a management VLAN, where the IPs in the subnet are used only for switches.

This setup explained is in a lab and this L3 switch is connected to a cisco router. Whenever machines sitting outside this router try to communicate to the lab, I am seeing inconsistent packet drops. It works sometimes and drops sometimes. I noticed that there are frequent spanning tree topology changes and thought it could be a spanning tree issue. I was using the RSTP default configuration, with L3 acting as the root node. Now, I have disabled STP on all the switches. Still intermittent packet drops are happening and unable to find out what is the problem.

Can anyone let me know what could be the problem? Are there any problems with the design created? Any pointer or help in this regard, will be really helpful.

[DaGuru]

Hi Prakash,

Can you revise your drawing to indicate where your redundant links are? Without them, spanning tree won't be adding any value to your architecture. Also, what is the spanning tree priority of your root switch? Perhaps, you could provide some sample switch configurations? Also, have you tired to identify which switches are acting as root? You should see the root bridge MAC address shift after each topology change.

If RSTP is in full default, your root switch may not be the one you think it is. Also, if all switches have the same priority, you may have two or more switches going through an election war and thus you'll see lots of topology changes and sporadic behavior.


Thanks
Dennis

[OLARU Dan]

PMS,
you mentioned you have packetloss even in the absence of spanning tree protocol. On which switch do you see the packetloss? - you may have to update the firmware, or you may have a faulty module or, worse, faulty backplane.

If you have redundant links between the switches, trunk them.

[OLARU Dan]

Daniel Kulinski has the same problem. Check his thread.

[P.M.S.Prakash]

Hi Dennis and Dan,

I have not upgraded the firmware to 10.61 like Daniel. But still was facing the problem with 10.55 firmware.

So, I replaced 5308 with 5400 as L3 switch and the packet drop was significantly improved, though the cpu started to shoot up due to spanning tree problem. But, as I had indicated in my diagram, I don't have any redundant links anywhere and the cause could be due to mistaken looping of cables at the machines. So, enabled RSTP on all the switches and it improved the situation. Still there was intermittent packet drops. After investigation found that if the switch links are created as trunk, the packet drop is not there. Initially, it was just tagged with all VLANs. It looks strange to me. It was working initially and suddenly started to create problem.

One question I have is on the topology counter. It keeps changing sometimes and I feel it is normal, as there are new connections in the form of switches or servers being added to the network, which results in these changes. Is that correct? Or the counter changing periodically is not good?

Best Regards
Prakash

[DaGuru]

In the case of adding new switches to the network, a topology change should occur as long as the new switch is configured to participate with the spanning tree. If that change count is abnormally high, it can be a sign that there are problems, such as two switches battling to take control as root, miss configurations, or perhaps a problem with physical link stability...
If, there are no redundant links in the overall architecture, they why are you running spanning three? Is this just to deal with network loops at the edge of your network?
Usually, RSTP will converge fast enough and the buffering technology in the switches should prevent packet loss from being unacceptable. If this is not the case, then it might be worth considering the idea of treating each edge switch as its own STP Domain. Iâ d like to see input from some of the other pros on here before I go fishing in that pondâ ¦ ï

[P.M.S.Prakash]

Hi Dennis,

I was told by one of the local HP procurve engineer that we should have RSTP enabled, even though I don't have redundant links in my network architecture. Use the default RSTP configuration on all the switches and make the core switch a root by reducing the priority. I believe that it helps in uncovering the redundant links at the edge switches. Even though my current design does not have redundant links, I plan to interconnect the distribution switches in the floors with GBIC links, so that it can be used for intra vlan switching packets instead of loading the core switch.

Also, I wanted to use RSTP to prevent loops created at the edge switches. I found that there is a loop-protection option for ports, which might help to prevent such loops bring down the network. Is that correct?

If that doesn't help, I will think about creating STP domains at the edge switch levels. Currently, the network is working well with STP enabled.

Best Regards
Prakash

[Andrew_291]

We have the network with two core 5308xl and many 4108gl and 8000m edge switches .All edge active devices are connected between neighbours with two links (optical line and "copper" trunk)and have one link to each core switch(by example, if edge 4108 have copper trunk to one of two 5308, then it has optical line to another).Core switches have two redundant links between.On all that network works STP, because 8000m cannot support RSTP.All works fine since 2004 year - no errors, no abnormal situations.Core switches have priority 128 and 256, edge switches - 1024 for 4108gl and 32768 for 8000m.All optical ports and copper trunks are tagged for all vlans, and each switch have the same set of static vlans (nothing GVRP at all !!!) We have not any broadcast control.

My opinion is that You may set STP on all switches on your network , set priority on each of them .And You MUST CLOSELY show vlans and links between switches.Seems to me that issue will be there.

[Andrew_291]

Wow .Another one. Try to set THE SAME SET OF VLANS ON ALL switches, as it write in docs 8)))
Hope it helps.