search

manuelkasper / AS-Stats

A simple tool to generate per-AS traffic graphs from NetFlow/sFlow records
BSD 2-Clause "Simplified" License
194 stars 67 forks source link

Incorrect Bandwidth on Juniper MX #42

Open GizmoV opened 9 years ago

GizmoV commented 9 years ago

Dear Manuel! Thank you for your work! Forced to turn to you for help, so as to solve the problem for a month I didn't get :(

We have 2 border routers of Juniper MX-series, one MX80 and one MX240. I tried to configure your scripts with each of them using Netflow v5/v8 and IPFIX and with the same settings got different results, and different from the indications by SNMP.

The results of MX240 like the truth at least 50-70% when the results of MX80 like no more than 20% of truth. Here is my current sampling settings:

sampling { sample-once; input { rate 100; max-packets-per-second 65535; } family inet { output { flow-inactive-timeout 15; flow-active-timeout 60; flow-server 192.168.90.100 { port 9000; autonomous-system-type origin; no-local-dump; source-address 192.168.90.91; version 5; } } } }

As I said - using Netflow v5/v8 or IPFIX - gives the same result. And here is graphs compare.

MX240 IP-transit 10G port.

telia

MX240 peering 10G port.

data

MX80 IP-transit 10G port.

retn

MX80 peering 1G port.

msk

I tried to change various settings of sampling rate as on the router and in your script, and your tips from the next issue: https://github.com/manuelkasper/AS-Stats/issues/4

Change $ascache_flush_interval = 10 gave a small positive result is higher-quality graphs, but to solve the problem of incorrect bandwidth I never got.

Asking for your help! I would very much like to use the product and send you a donation.

JackSlateur commented 9 years ago

You are misunderstanding graphs ! On the left, SNMP-based, you've all the data On the right, the graph show you the data consumed by the 10 "hungrier" ASN.

So, on the right, you have thousands ASN that are not drawn :)

Depending on your kind of data type, you may have a lot of ASN which consumes "almost the same bandwidth", that might explain such graphs.

For information, in my compagny (ISP), on IXP like AMS-IX, the diff is 40%

GizmoV commented 9 years ago

Dear clobrother! I can't fully agree with you :) At least in question with MX80. Look at the comparison №3 at the junction with the RETN. If the right really 10 most hungry AS something they can use only 10% of the strip to the left? Also I forgot to add that MX80, depending on the time of day - almost not falling and increase bandwidth traffic. Thank you for your help, me in had no idea :) But I think the problem is in something else too.

bpbp-boop commented 3 years ago

this is an old question, but I believe the magic command you're looking for is flex-flow-sizing under the fpc you're using.

reference: https://www.juniper.net/documentation/us/en/software/junos/flow-monitoring/topics/ref/statement/flex-flow-sizing-edit-chassis.html