How to implement Fast loop detection?

Unanswered Question
Aug 31st, 2006
User Badges:

Hello,

I recently set up my first, very simple spanning tree network, consisting of one RSTP-capable switch, and several other simple un-managed switches. The topology is as follows:


.......HOST

...........|

.....Managed SW

....../.......\

.....|......switch1 ---- Server1

.....|.........|

.....|......switch2 ---- Server2

.....|.........|

......\___/


(Sorry for the crappy ascii graphic)


The main goal is to be able to connect it in a loop, and have have connectivity btwn HOST and servers, then break the loop anywhere, and have it continue to function, with as short of a communication break as possible. It works, but comms are lost for 12-15 seconds, down from 45sec after tweaking all timeouts to their minimums. Anybody have an idea improve this recovery time? We'd like a recovery time of less than 2 seconds for our system to work properly.


My current settings are as follows:

Global settings:


- STP: Enabled

- Version: RSTP

- Hello time: 1 sec

- max age: 6 sec

- Forward Delay: 6 sec

- Priority: 32768


Settings for ports 1-24:

- P2P: Yes

- State: Disabled (turns off STP for these ports)

Settings for ports 25,26:

- P2P: Yes

- State: Enabled

- Priority: 128

- Edge: No

Ports 25 and 26 are the fiber ports, where the primary and backup fibers connect.


With these settings, it takes about 14 seconds to recover from a topology change.


My question is, can I use one of Cisco's features such as Uplink Fast to improve this performance to < 2 sec, and if so, will it work with just one managed switch?


We have a strong need to keep all of the other switches unmanaged, except the top one that actually detects the ring or break. The other switches are under water and combined with other electronics, and would be expensive to modify or replace with managed type switches.


Thanks!

Randy



  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
jackyoung Thu, 08/31/2006 - 17:30
User Badges:
  • Gold, 750 points or more

Thanks for the great diagram. It is earier for reading.


If you use RSTP, it already improved the convergence time. However, are you sure that all equipment are config. as RSTP ? Will any of the switch still config. as traditional STP ?


In normal cases, RSTP is good enough and no need uplink fast, portfast, backbone fast STP from Cisco.


If you really require < 2 sec fall-over, you may consider to implement layer 3 routing to use routing protocol instead of STP, and the design also need to change.


Or you can redesign the network hub-and-spoke design then enable Etherchannel (w/ 2 uplinks) between each spoke to the hub switch. Then two uplinks will be load-sharing and provide resiliency even one link down w/o require STP process.


Hope this helps.

Actions

This Discussion