[j-nsp] telemetry analytics - mx960 - npu packet rate concerns

[Aaron Gould]

Using my JTI/Chronograf/Grafana web interface I'm trying to understand some
of the telemetry analytics data I'm seeing coming from what appears to be
the sensor resource of my MX960 corresponding to
/junos/system/linecard/npu/utilization/ .. The field seen on chronograf that
I'm watching is "npu_util_stats.packets.rate"

 

When using the Chronograf data explorer and picking one MX960 and a certain
_seq number (0-14 , I don't know what these are) I'm seeing some significant
drops in the graph during peak time (approx. 7 - 10 p.m.) watching
"npu_util_stats.packets.rate" with mean function (as opposed to median,
count, min, max, etc, etc).  In other words, the graph shows a typical
ramp-up approaching peak times, and ramp-down during the late night hours..
But about a week ago, I started seeing dramatic drops/sags in the graph
during those 7-10 p.m. hours.

 

I'd like to try to figure out what those drops are related to. I'm wondering
if this is the MS-MPC-128G npu's in-use for my cgnat.. I've been loading it
up quite a bit lately with thousands more subscribers behind it, and am
trying to watch how it scales. and if I have any reason for concerns
regarding resource load, etc.

 

(should I post this to NANOG also?)

 

-Aaron