Alex:
There is a known (at least to networks with large Cascade deployments)
problem with Ciscos and Cascades in that high link utilization will cause
LMI drops and thus interface drops. The problem as was explained to me is
that LMI is not prioritized over other traffic, so when discards have to
occur, LMI can also go. This has been known to occur as low as 60-70% line
utilization.
So far the only resolution to this is "no keepalives". You now have to
rely on the routing protocol to determine when the far end is unreachable
(i.e. link is down).
I haven't seen this occur on HDLC links, except when there were sufficient
BGP bounces/instability to max out the CPU and cause further BGP bouncing.
...karl
At 06:59 PM 3/28/98 -0500, you wrote:
>I posted about this before in the context of FR over HSSI and Greg Ketell
>posted some useful stuff back, but the problem is wider and repeatable.
>I don't know if anyone else has seen this.
>
>7513 & 7507, running 11.1.12CA1, 11.1.15CA1 and 11.1.17CA1 with CX-HIP,
>both under point to point HDLC through Digilinks, and running FR
>into a Cascade FR switch, appear to drop line protocol when the router
>hits high CPU load (for instance through oscillating lines or BGP
>recalculation). As this causes oscillating lines and BGP recalculation,
>this is a bad thing.
>
>Has anyone else seen this? (Bug Navigator hasn't). Does anyone have a fix?
>I have increased the broadcast queues and played with various LMI timers
>for the FR stuff which mitigated the problem a little on the FR side.
>It appears to be perfectly repeatable (unfortunately today gave me the
>chance to test this...)
>
>--
>Alex Bligh
>GX Networks (formerly Xara Networks)
>
>
>
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:16 EDT