[j-nsp] Limit on interfaces in bundle
Mark Tinka
mark.tinka at seacom.mu
Fri Oct 30 12:20:58 EDT 2015
On 30/Oct/15 14:44, Michael Hare wrote:
> All-
>
> You are correct about ECMP, I encountered this 12+ months ago and forgot that specific details. Mark said he put the whole AE bundle into per packet mode [on a short LAN hop], which can also cause reordering [not just the LSP] if used incorrectly.
We've been tracking this for a couple of months now (per-packet
spraying), and no issues to report, really.
The thing to remember is that if policing is done on a LAG interface,
without equal loading of traffic across all members of the link, you
would end up with a fraction of the throughput on that interface as
Junos equally applies the policer across all member links of the LAG.
With two member links, a bias of the hashing on one member means only
50% of the traffic gets through the policer. If three member links, 33%
of the traffic gets through, e.t.c.
So the per-packet mode has really helped here.
Mark.
More information about the juniper-nsp
mailing list