Hey everybody,

The other thing I'm working on with Juniper flows are "long flows", we
have various flow timeouts set to sane values, but keep seeing really long
flows arrive at our collection station, which is messing with certain
analysis programs.

I wrote a little script that shows us how many flows and bytes from a
given flow file are going into what time buckets, where I've set the
buckets to be 0-900 seconds, 900-1800, etc.  In my dream world no flow
would be reported to last over 900 seconds (and there would be free It's
It Ice Cream Sandwiches).

Here is a "random" sample:
(inr-001 and inr-002 are m40e's, inr-003 and inr-004 are m7i's)

netflow.inr-001/ft-v05.2006-08-11.040000-0700:
0 111140 6006217920

netflow.inr-002/ft-v05.2006-08-11.040000-0700:
0 164664 17977073440

netflow.inr-003/ft-v05.2006-08-11.040000-0700:
0 711499 3435343092
900 2615 872297513
1800 881 261323019
2700 263 44570234
3600 79 30430681
4500 51 4852349
5400 21 3418594
6300 11 7041484
7200 4 84925
8100 2 7429

netflow.inr-004/ft-v05.2006-08-11.040000-0700:
0 412570 1263148945
900 592 759589358
1800 213 101194786
2700 89 66340933
3600 47 40617598
4500 31 1575887
5400 12 1558969
6300 7 2189981
7200 5 42184
8100 4 141120
9000 1 5639631
11700 1 38076
16200 1 24810

I've seen much longer, including **54 hours** :)

We got a "promising" response recently from their TAC, which hinted at the
idea that the beginning time of a flow is *not* reset even when it gets
"flushed" and set off in a PDU:  Instead, the counters are reset to zero
(and presumably the end time changes) but the flow remains in some sense.
We're not completely sure that we are interpreting their comments right,
so we're following up.

Mike
_______________________________________________
Flow-tools mailing list
[EMAIL PROTECTED]
http://mailman.splintered.net/mailman/listinfo/flow-tools

Reply via email to