Ticket #251 (closed defect: fixed)

Opened 6 years ago

Last modified 6 years ago

mesoscale loop

Reported by: ibaldin Owned by: ckh
Priority: major Milestone:
Component: Infrastructure: NLR links Version: baseline
Keywords: Cc: tupty@…

Description

Hi RENCI folks,

Starting last Friday around 2:10 PM EDT, we started seeing elevated
broadcast rates in the shared mesoscale VLAN. After collecting some
data and trying various things out, we discovered that if we cut the NLR
connection to RENCI out of slices participating in the mesoscale VLAN,
the broadcast counters returned to normal.

It seems like there might be something like a broadcast loop somewhere
downstream of the NLR connection facing RENCI in the shared mesoscale
VLAN. Can someone at RENCI work with us to try to find where this is
happening?

Thanks,

Change History

Changed 6 years ago by ibaldin

  • cc tupty@… added

Changed 6 years ago by ckh

The potential for a loop was created by connecting two OpenFlow? enabled ports on the 8264 to each other (in preparation for hybrid mode). The mesoscale Vlan-1750 controller receives a list of all 8264 ports from FOAM/Flowvisor including the two ports that are connected back-to-back. The mesoscale controller sends arp requests to the full list it receives from FOAM. The back-to-back ports facilitate a flood of arps that recirculate themselves through each other and the upstream port facing BBN. Removing the back-to-back cable and restarting the controller corrected the problem.

When hybrid mode is implemented, it will connect a traditional Ethenet port to an OpenFlow? port to establish connectivity between the two halves of the switch. The list of ports that FOAM/Flowvisor send to the controller will only contain the OpenFlow? port. So it shouldn't present a problem.

Changed 6 years ago by ckh

  • status changed from new to closed
  • resolution set to fixed
Note: See TracTickets for help on using tickets.