[c-nsp] Randomly Dead Sessions with MLPPP(oE)

Anton Kapela tkapela at gmail.com
Thu Oct 4 22:13:23 EDT 2012


On Sat, Sep 29, 2012 at 4:41 PM, Gary T. Giesen <giesen at snickers.org> wrote:

> I've got 2x 7206 VXR/NPE-G1 running 12.2(33) SRE6 acting as LNSs being
> passed PPPoE sessions from an incumbent provider via L2TP. A couple
> customers (including my personal Internet connection) using Multilink
> PPPoE (over ADSL) are randomly experiencing dead sessions several
> times per day. All 3 sessions will stay up (2x link + bundle session)

I'd recommend experimenting with the following parameters:

 ppp multilink fragment size <xxx> (try values from 500 to 800 bytes,
could be tx or rx reassembly bug)
 ppp multilink fragment disable (perhaps it's frag processing that
rolls counters over a cliff -- just force per member round robin'ing
with this)

also for grins, I've seen cases where mlppp frags on multi-member
links wedge the receiver of the frames when the DF bit is 1. try
wiping it on your IP transit facing ports, and at the user facing CPE
ethernet port with something like:

route-map clear-df permit 10
 set ip df 0

Lastly, do you enable or use mlppp interleaving, per chance?

-Tk


More information about the cisco-nsp mailing list