In that case, another Faloutsos paper would be of interest:

2002 Performance - best student paper award: Mengzhi Wang, Anastassia
Ailamaki and  Christos Faloutsos, *Capturing the spatio-temporal behavior of
real traffic 
data<http://www.cs.cmu.edu/~christos/PUBLICATIONS/performance02.pdf>
* Performance 2002 (IFIP Int. Symp. on Computer Performance Modeling,
Measurement and Evaluation), Rome, Italy, Sept. 2002

http://www.cs.cmu.edu/~christos/PUBLICATIONS/performance02.pdf

I don't see that these techniques will necessarily scale, but violation of a
traffic model might well be a good anomaly detector for your problem.

On Sun, Oct 3, 2010 at 6:57 AM, Latency Buster <[email protected]>wrote:

> >That's a nice non-answer.  The term "network data" and "identify some
> > definite patterns" doesn't say enough to answer anything specifically.
>  If
> > you know what kind of patterns you are looking for, or if you could say
> what
> > the data contains, you could get
> > some help here.  With open source, you get back much more if you give a
> > little to start with.
>
> Thanks for the pointers. Fuzzing  was not my intension but I was
> fearing that I might coalesce two different topics under one heading.
>
> Now,to provide some clarity around this: We have around 1TB of data as
> the ultimate aim but with 50GB of data to start with. The data
> consists of identifying a definite pattern that arises when people are
> making calls. We are a call center using IP phone. There are certain
> times of the day when not only the volume but the encoded speech
> patterns contains a certain overlap in their freq spectrum. So, our
> aim to is cluster/extract some representative feature sets of the
> decoded voice signals (or signal fingerprinting), correlate with the
> network feature sets (say ratio of 3sigma after (0,1) normalization)
> and check if there exists some useful information around these.
>
> Thanks,
>

Reply via email to