This is a trace for my darknet and opennet node both running ... also,
it includes incoming packets as well as outgoing.

Still:
Total small packets: 321,428.

This is 267 small packets per second! Lets divide it by 2 for 2 nodes,
we get 130 packets per second.

The message most frequently sent is FNPLinkPong. The code will send no
more than one message every 200 milliseconds (total across all nodes),
so in theory we shouldn't have more than 5 per second outgoing per
node... of course, that is FNPLinkPing's. Since I have 23 active
connections, most of which will have relatively few connections, it is
quite possible that they would send me relatively many FNPLinkPing's per
second... but still, this cannot account for the bulk of it because there
are only twice as many FNPLinkPong's as packetTransmit's in a dump I
made of what messages are being sent... (1kB+ packets are usually
packetTransmit's, as packetTransmit's are always 1kB).

Note also that this is a very noisy sample; there may be others using
the link, and I can't see immediately how to only capture outgoing UDP
in iptraf.

Suppose the average connected peer had 5 other peers. I have 23 peers.
Each would therefore send me one packet per second. That still doesn't
even begin to explain the situation... hmmmmm.

On Thu, Apr 06, 2006 at 10:30:39PM +0100, Matthew Toseland wrote:
> Packet Size (bytes)      Count     Packet Size (bytes)     Count
>     1 to   75:           20248      751 to  825:              31
>    76 to  150:          239541      826 to  900:              28
>   151 to  225:           52956      901 to  975:              35
>   226 to  300:            7297      976 to 1050:              15
>   301 to  375:             547     1051 to 1125:              35
>   376 to  450:             385     1126 to 1200:           17657
>   451 to  525:             216     1201 to 1275:             129
>   526 to  600:              44     1276 to 1350:              97
>   601 to  675:              36     1351 to 1425:             139
>   676 to  750:              14     1426 to 1500+:           1311
> 
> This is a log of packet size from my node over a period of 20 minutes.
> Actually of my 2 nodes, and probably a little TCP traffic too, but not
> much.
> 
> Interesting features:
> 
> 76-150-byte packets: 239451 * 100 = 23,945,100
> 1126-1200-byte packets: 17657 * 1150 = 20,305,550
> 
> Of the first group, 56 bytes per packet is overhead, so 13,409,256 bytes
> overhead out of that 23MB - something like a quarter of the whole.
> 
> Obviously having many variable sized small packets is a bad thing for
> security, but surely it is a good thing for latency to be as low as
> possible by sending messages immediately?
> 
> Another interesting point: If more than half of our bandwidth usage is
> on small packets, then our policy of only bandwidth limiting large
> packets cannot possibly work.
-- 
Matthew J Toseland - toad at amphibian.dyndns.org
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: 
<https://emu.freenetproject.org/pipermail/devl/attachments/20060406/b441de9f/attachment.pgp>

Reply via email to