pmacct-discussion  

Re: [pmacct-discussion] reloading config & accuracy

Tony
Mon, 21 Sep 2009 17:19:26 -0700

Hi Paolo

--- On Mon, 21/9/09, Paolo Lucente <pa...@pmacct.net> wrote:

> Hi Tony,
> 
> On Sun, Sep 20, 2009 at 06:03:18PM -0700, Tony wrote:
> 
> > I haven't upgraded yet, I will be doing that now, but
> I wanted to give you some feedback on what I'm seeing in the
> old version and we can see if it persists to the new
> version.
> >
> > [ ... ]
> 
> To confirm you that that behaviour was precisely target of
> a
> bugfix which was issued a few weeks ago to the CVS and now
> is
> already part of 0.12.0rc2 release. Let me know how it goes
> after the upgrade.
> 
> Cheers,
> Paolo
>

I've upgraded to the latest 12.0rc2 version and the results are a lot better. 
An example of the data is:

 5,716,873,659   5,615,718,189  101155470       1.801%
 5,565,297,967   5,415,906,647  149391320       2.758%
 2,383,851,454   2,324,030,264  59821190        2.574%
 1,963,615,214   1,890,286,868  73328346        3.879%
 1,928,935,113   1,870,256,519  58678594        3.137%

Where again, first column is data after adjb=26, second column is without any 
adjustment, third column is difference between first two and the fourth column 
is percentage difference.

Each line of data is the total for a single IP address over the period from 
when I upgraded yesterday to just now (about 20 hours). This is showing the 
kind of results I expect.

To convert the above to average MTU size of the packets, it would be like this:

1.8% = 1444 bytes
2.5% = 1040 bytes
3.8% = 684 bytes

So a value of 3.8% overhead would be about as high as I'd expect it to go 
otherwise the average MTU across all packets is getting too low which shouldn't 
really happen (except for special cases).

My above samples are from the highest bandwidth users, the less data there is, 
the higher the percentage overhead appears to be (probably due to more small 
background traffic, like random pings, etc). This is also less than one days 
worth of data so it should get better and I hope move more towards a consistent 
<3% overhead over time.

Now that adjb seems to be doing what it is supposed to do I will accumulate a 
few more days/weeks of data and compare the values from pmacct (with adjb) to 
those being recorded directly by the packeteer. Hopefully they will be a lot 
closer now.

Thanks for all your help so far.

regards,
Tony.



      
__________________________________________________________________________________
Get more done like never before with Yahoo!7 Mail.
Learn more: http://au.overview.mail.yahoo.com/

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