[j-nsp] ex4500 best-effort drops nowhere near congested

[ryanL]

hi list.

i'm guessing this is a buffer thing, but i can't explain why it only
happens on my 1ge ports and not when i punt the traffic over an 10ge
port.

i have two 1ge in an ae bundle. both of them basically look like this:

  Statistics last cleared: 2013-05-02 00:16:50 UTC (00:02:02 ago)
  Traffic statistics:
   Input  bytes  :           2584384859            168276320 bps
   Output bytes  :           4108072542            289189088 bps
   Input  packets:              4003290                33943 pps
   Output packets:              6080350                56645 pps

  Egress queues: 8 supported, 4 in use
  Queue counters:       Queued packets  Transmitted packets      Dropped packets
    0 best-effort                    0              6081024                  526
    1 assured-forw                   0                    0                    0
    5 expedited-fo                   0                    0                    0
    7 network-cont                   0                  591                    0
  Queue number:         Mapped forwarding classes
    0                   best-effort
    1                   assured-forwarding
    5                   expedited-forwarding
    7                   network-control

i have fairly steady state traffic, maybe a little bursty here and
there? but again, if i redirect this traffic over a 10ge port i never
see a single drop.

jtac suggested: set class-of-service shared-buffer percent 100

this helped reduce the dropped packet count, but i'm still confused as
to why i see any at all on my 1ge ports and not when i punt it over
the 10ge. is buffer allocated on the pfe differently based on what
transceiver you are using?

other than the above cos command, i've made no changes to default cos
configuration on either of my 4500's, which both see the same problem.
on the flip side of these 1ge connections are MX5's, which never
register any packet drops at all.

i guess one thing to realize is that the 1ge interfaces are routed
(bgp to my MX5's), but so is the 10ge interface, and the forwarding
should all be processed in hardware anyways.

ver: 12.2R2.4

appreciate any insight!

ryan