Question about 2900xl and 5505 switches

[Stupendo]

We have two 2900xl switches in the same comms room and both of the switches have a fibre link that goes back to our 5505 switch in another comms room.

We would like to interconnect the two 2900xl but everytime we try this it just disables the port between the two 2900's.

Does anyone know how we can succesfully interconnect the two 2900xl to allow traffic to flow between them and not just via the 5505?

Hope you can all understand this if not by all means ask me more questions!

Cheers,

Jon.

 

[wybnormal]

Can we say spanning tree? When you interconnect the two 2900s, you create a redundent link between the two. Your primary link is the link back to the Cat5K and spanning tree prevents the type of data loop that are trying to create.

The first question is why? what is wrong with feeding the 2900s back to the big cat? This needs to be addresses first. Disabling spanning tree is not a real option as you will end up with connections that go away, break, flood with broadcasts and a host of other nasties. It is possible to force the path by adjusting certain spanning tree parameters.. but Cisco does not recommend this unless the engineer doing this has a very good grip on what he/she is doing.

See this linke
how spanning tree works

http://www.cisco.com/univercd/cc/td/doc/product/lan/cat6000/sw_6_3/confg_gd/spantree.htm

MikeS
Find me at

http://www.packetattack.com

"Take advantage of the enemy's unreadiness, make your way by unexpected routes, and attack unguarded spots."
Sun Tzu

 

[Stupendo]

Sorry for being a complete dumbass and actualy not giving you all the information.
We actually have one NT server on one of the 2900's running IP and one netware box on the other 2900 running IPX. The IP traffic seems to have no problem going up to the 5505 and back down to the other 2900 but the IPX traffic does seem to suffer.

We have to resort to interconnecting the two 2900's and unplugging one of the links back to the 5505 to help us with the netware speed problems.

We cannot move either of the server's and at this moment in time we cannot get rid of the netware server.

 

[wybnormal]

When you say IPX suffers.. what do you mean? slow response? SAP list losing servers? Which flavor of Netware are we talking about here? ancient 3.12? 4 or 5.x? Which ethernet frame type? 802.3? 802.2? Are the ports on the netware server locked to a specific speed/duplex? I remember that the Compaq Flex nics had issues running at 100/full with IPX.. effectively it was too fast for the server which needed a bit more time to process the SAP packets correctly. We ended up with servers dropping in an out of the server list. This also shows up on routers if you have fast switching enabled instead of process switching. It's fixable but it can drive you nuts in the process.

There are probably a few ways this can be handled with just a bit more detail.

MikeS
Find me at

http://www.packetattack.com

"Take advantage of the enemy's unreadiness, make your way by unexpected routes, and attack unguarded spots."
Sun Tzu

 

[adam108]

wyb is certainly on the right track for you. To which switch are your IPX consumers who are having problems connected, the 5500 or the non-IPX 2900. If they are connected to the 2900 you may need to configure your spanning tree to ensure that the ipx-owning 2900 is the designated port for this segment. If your users branch from the 5500, determine which port on the ipx-2900 is the root port (i.e. direct to 5500 or via non-ipx-2900 to 5500. I.e., where does the traffic flow when you have unhooked one of the 2900's, which is desirable vs when both 2900's are hooked into the 5500?

 

[Stupendo]

To answer some of the questions, we are seeing slow response time's to applications on the netware server, if the user concerned has to travel up the fibre from one 2900 to the 5505 and back down the fibre to the other 2900 with the netware box on it's very slow. If you connect the two 2900 together thus cutting out the 5505 its fine.
Netware version is 3.2 (old legacy system!), frame type is only 802.3 and the server is a HP LC3 with a 3com Fast Etherlink XL NIC, this is not locked down either on the server end or the switch end.

Hope this helps and thanks for all the advice so far.

Jon.

 

[wybnormal]

Now.. for the hundred dollar question

What version of IOS does the cat5K *AND* which type of blade are the 2900s plugged into? I found that some of the older 10/100Mbps blades have troubles with IPX when chained to switches. Again, it's a speed issue of the switch. One thought is to check error rates on the ports. The Cat5K uses store and forward switching. If there is an error, then the packet is dropped. It's not very forgiving in this regard. You could as an experiement, change it to cut-through switching which sends the packet on it's way regardless of *errors* real or not.

A Sniffer would be a great use here. You would see exactly what is happening to the packets. You would set up a span port and then mirror the problem fiber port to the spanport. Feed that into the Sniffer and see what is up.

You said the ports are not locked? Cisco's own best practices is to lock the ports to whatever speed/duplex they can handle. That is something easily done but be warned it will reset the port so dont do it in the middle of the day.

MikeS
Find me at

http://www.packetattack.com

"Take advantage of the enemy's unreadiness, make your way by unexpected routes, and attack unguarded spots."
Sun Tzu