On Thu, Jul 27, 2000 at 10:41:20AM -0700, Mark Milhollan wrote:
Bill Woodcock writes:
I'm interested in seeing the distribution of packet sizes across a 1500-byte-constrained measurement point, with "real internet traffic" going past (for some reasonable interpretation of that phrase).
Protocol Total Flows Packets Bytes Packets Active(Sec) Idle(Sec) -------- Flows /Sec /Flow /Pkt /Sec /Flow /Flow Total: 758613841 176.6 2 1170 440.8 6.1 14.6
I.e., on cisco routers with "Flow switching" enabled on all interfaces you want data included from, the router will give you that ...
show ip cache flow IP packet size distribution (1242M total packets): 1-32 64 96 128 160 192 224 256 288 320 352 384 416 448 480 .001 .436 .050 .018 .012 .008 .006 .005 .004 .005 .004 .006 .004 .003 .004
512 544 576 1024 1536 2048 2560 3072 3584 4096 4608 .003 .003 .071 .030 .318 .000 .000 .000 .000 .000 .000 [...]
Thanks (and to all others that replied). I'm actually looking for more granular data than this; I was hoping to map the precise overspill of AAL5 + SNAP + IP packet sizes modulo 48 bytes to help produce (with the cell header tax) an estimate of the cost of ATM vs POS on an expensive trans-pacific STM-1. Apparently hard data is easier for people to believe than my handwaving :) I don't _think_ that mean packet sizes will give me what I want (unless the spikes in the histogram at 552, 576 and 1500 bytes are large enough to render the rest of the distribution practically meaningless). However, my statistics skills lie somewhere on the poor side of crap, and if someone can point out the glaringly obvious way to derive the fixed-size-cell overhead from data such as that quoted, I'd be glad to hear it. Joe