Hi Jaromir,

On Sun, Jul 24, 2011 at 02:04:59PM +0200, Jarom?r ?ervenka wrote:

> In high traffic ( > 200 Mbit / sec.) pmacctd doesn't count all bytes and 
> packets. I found out that this happened when I use pretag file. I have pretag 
> map file with almost 3000 filters inside, for recognizing traffic which heads 
> outside the country. 

[ ... ]

> I also tried to compile pmacct against PF_RING libpcap, but it didn't help. I 
> also tried to "play" with different buffers in configuration file, but it 
> didn't help either. Do you have any idea why pmacctd doesn't count all 
> traffic in such conditions?

Then i would imagine you can spot CPU peaking at 100% ? Evaluating 3K
rules against maybe ~34K pps (ie. 200Mbit @ ~750 bytes/packet) can be
intensive. Can suggest a couple of things:

* As Slava pointed out, pre-process raw traffic and evaluate rules in
  nfacctd against NetFlow datagrams. Pre-processing can be done via an
  external software or with pmacct - by using the nfprobe plugin.

* Don't know what is the actual content of your tag rules but they are
  evaluated in the order you loaded them, just like you would expect of
  a set of firewall rules. Perhaps you can sort them placing the "most
  likely to be matched" rules first - and see if this is beneficial?

Cheers,
Paolo

> 18. 7. 2011 v 13:48, Paolo Lucente:
> 
> > Hi Jaromir,
> > 
> > Thanks for your interest into the pmacct project. Please find below
> > answers to your questions:
> > 
> > 1) Did you compile the package with --enable-ipv6 ? 
> > 
> > 2) Is it possible it's all traffic from the outside to IP addresses
> >   assigned to you but not used as a result of a scan? Or you see
> >   unused IP addresses generating traffic?
> > 
> > 3) 0.0.0.0 is traffic to/from some networks which are not listed as
> >   part of your networks.lst file. This is also briefly explained in
> >   the CONFIG-KEYS, networks_file part: "[ ... ] a) it allows to
> >   rewrite as zero IP addresses not included in any defined network
> >   range (ie. to avoid IP addresses external to the local domain to
> >   be accounted for)"
> > 
> > Cheers,
> > Paolo
> > 
> > On Fri, Jul 15, 2011 at 11:57:41AM +0200, Jarom?r ?ervenka wrote:
> >> Hello to all,
> >> 
> >> first I must say, that pmacct is great piece of software, thanks for it.  
> >> May I have few questions for more advanced user, than I am?
> >> 
> >> Let's start with my current configuration, for one server:
> >> 
> >> pmacctd.conf: http://paste.opensuse.org/97918056
> >> networks.lst: http://paste.opensuse.org/88433187
> >> pretag.map: http://paste.opensuse.org/31181373
> >> 
> >> 1) First question is regarding to IPv6 i pre_tag_map file. When I put  
> >> IPv6 network inside, I get this error:
> >> 
> >> INFO ( default/core ): Trying to (re)load map: /etc/pmacct/pretag.map
> >> ERROR ( /etc/pmacct/pretag.map ): malformed filter: unknown network 'a01'
> >> Line 2 ignored.
> >> INFO ( default/core ): map '/etc/pmacct/pretag.map' successfully 
> >> (re)loaded.
> >> 
> >> Is it possible to use IPv6 networks inside pre_tag_map file? I would  
> >> like to distinguish IPv4 and IPv6 traffic.
> >> 
> >> 2) As you can see, I've configured pmacctd for accounting total traffic  
> >> for IN and OUT (in two different SQL tables), only for my two networks -  
> >> one is IPv4, second one IPv6. But I have just few IP addresses allocated  
> >> from these networks for my servers (for exapmple: 80.79.27.58-  
> >> 80.79.27.71). But the results shows me, that there is some communication  
> >> from my network to the internet from addresses, which are not assigned  
> >> any of my servers or ethernet cards (like .83 , .76, etc.) . How is it  
> >> possible?
> >> 
> >> 3) My last question is regarding to aggregation total IN / OUT of  
> >> networks. In the results there is network 0.0.0.0 which has significant  
> >> nr. of packets and bytes. What does it mean?
> >> 
> >> Thank you for any advice,
> >> Jaromir Cervenka


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

Reply via email to