It seems that over time, bgp updates on a router with full routes
will pretty thoroughly fragment memory. Whether the problem is
a failure to recover and coalesce the free space or it's just the
structure of the bgp routing table entries that tends to leave one
lots of interspresed active and deleted entries is unclear.
Doing a clear ip bgp * will usually free up some bigger chunks
in an emergency siutation and may or may not be less disruptive
of your network than a "reload".
> From: Dave Curado <email@example.com>
> To: firstname.lastname@example.org
> Subject: [nsp] largest free block with uptime >1 year
> X-Mailing-List: <email@example.com> archive/latest/3056
> X-Loop: firstname.lastname@example.org
> Precedence: list
> Resent-Sender: email@example.com
> We saw a signficant routing flap tonight, that resulted in some
> of our ebgp speaking routers running out of memory.
> On closer inspection, the routers all have enough memory
> (128M, with 40M or so free) but the largest free contiguous
> block is very small on routers that have been up for a year
> or so. Has anyone seen this?
> I've always believed that cisco routers do not require a
> "theraputic" reboot, but the data I'm looking at now suggests
> otherwise. (I rebooted one of these routers that had run out
> of memory, and it's largest contiguous block is much larger
> Here's one of the more unhealthy routers that ran out of memory:
> show mem
> Head Total(b) Used(b) Free(b) Lowest(b) Largest(b)
> Processor 60AD9EE0 122839328 82389772 40449556 145788 4194368
> Fast 60AB9EE0 131072 115000 16072 16072 16028
> router uptime is 43 weeks, 17 hours, 46 minutes
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:13 EDT