Hi Paolo,

1. Ok! Indeed some specific doc for pmacct would be great but there is already 
public doc (i.e. specific for Snort) that should be really easy to use for 
pmacct.

2. I don't get it. If I'm right, it is pmacct who receives traffic directly 
from network (through libcap) and generates the flows, so isn't pmacct who 
generates these fields? If I do a tcpdump on the interface I can see resolution 
in ms. However I did set "timestamps_secs: true" and now in nfsen it reports 
start time at linux epoch (1970-01-01) and the duration reports it takes until 
now (i.e. 1413383398.940 seconds).

NFSEN Example
Top 10 IP Addr ordered by flows:
Date first seen          Duration Proto           IP Addr    Flows(%)     
Packets(%)       Bytes(%)         pps      bps   bpp
1970-01-01 01:00:00.000 1413383398.940 any      xxx.xx.xxx.xxx    3.2 M( 9.0)   
10.7 M( 1.9)    2.2 G( 0.5)        0       12   204
1970-01-01 01:00:00.000 1413383398.992 any      xxx.xx.xxx.xxx    2.7 M( 7.5)   
65.0 M(11.6)   61.9 G(13.0)        0      350   952

3. I don't fully understand the usage of 'aggregate_primitives'. Can I see a 
real example somewhere?

Thank you very much,
Xavi

-----Mensaje original-----
De: pmacct-discussion [mailto:[email protected]] En nombre 
de Paolo Lucente
Enviado el: dimecres, 15 d'octubre de 2014 10:47
Para: [email protected]
Asunto: Re: [pmacct-discussion] Using pmacct instead nprobe

Hi Xavier,

To your questions:

* No, you can't configure the amount of threads. pmacct uses coarse-grained
  multi-threading meaning specific functions, ie. BGP or IGP daemons, are
  embedded in a separate thread. Should you want to scale beyond a single
  core, you can use PF_RING as a "load-balancer" to multiple pmacctd's. I'm
  no expert of PF_RING but this has been done and, if you are interested, i
  can fetch the required info for you (maybe good idea to build some basic
  public doc in this sense).

* About the specific FIRST_SWITCHED and LAST_SWITCHED issue: this is maybe
  because you receive times in msec resolution instead? ie. you receive field
  types #152 and #153 instead of #21 and #22? If this is the case you can
  revert to secs counters by adding to your config: "timestamps_secs: true".
  You can control the primitives via the 'aggregate' directive; finally you
  can define your own primitives via the 'aggregate_primitives' directive
  (much has been done here already but much is yet to come and should you
  have any feedback please let me know).

Cheers,
Paolo

On Tue, Oct 14, 2014 at 09:37:38AM +0000, Xavier Romero wrote:
> Hello,
> 
> I just discovered yesterday pmacct by doing some google search and I wanted 
> to give it a try.
> Our scenario is pretty simple: We have a pair of Linux boxes receiving port 
> mirrors from our network COREs which we use to generate netflows with nprobe 
> and visualize them with nfsen and Kibana.
> 
> We replaced nprobe with pmacct and a very simple configuration:
> daemonize: true
> interface: eth2
> aggregate: src_host, dst_host, src_port, dst_port, proto, tos
> plugins: nfprobe
> nfprobe_receiver: 10.60.1.69:9970
> nfprobe_version: 9
> pidfile: /var/run/pmacctd-eth2.pid
> syslog: daemon
> 
> I'm so impressed with the performance, since it's being much less 
> cpu-intensive than nprobe. But I've some doubts:
> 
> *         There is any way to configure thread number or something like this? 
> I've enabled threads at compile time but I always see just 2 threads.
> 
> *         Can I configure which fields are being informed in the generated 
> flow? By using pmacct instead nprobe I realize that I'm missing some fields, 
> i.e. FIRST_SWITCHED and LAST_SWITCHED. I see how can I define flow 
> aggregation and fields that can be used (-a) but not how to define which 
> fields will be sent in the flow.
> 
> Thank you!,
> Xavier Romero

> _______________________________________________
> pmacct-discussion mailing list
> http://www.pmacct.net/#mailinglists


_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists

_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists

Reply via email to