Estimation of future VM requirements

Gangadhar NPK · science forum beginner Joined: 13 May 2006 Posts: 1

All,
I have a problem that I am trying to solve, and am guessing that there
must be a statistical method to solve this.

The problem:
To be able to estimate the future memory requirements of an
application, given data of the current consumption of VM of the
application. i.e., I gather the memory usage of the application for a
considerable time, and then like to predict what will be the memory
needed in the future.

Description:
There is an application comprising of lots of sub-applications which
are continuously running on a server. These individual applications are
high transactional servers, and they run on high end machines. I have
the data of the memory consumption of the servers for a certain period
of time (say a day), taken at considerable intervals. With this data,
one needs to estimate the memory requirements in the future. This will
be useful to calculate sizing requirements of the hardware, given a
certain amount of business transactions happening. I don't think we can
linearly estimate the memory requirements, as the usage of the memory
is different at various times of the day and nor is it random, as there
is a certain pattern in the business usage of
the application(s).

What is desirable is to find what kind of distribution this data fits
into and then use the distribution characteristics to estimate the
future consumption. I checked with a few knowledgeable people in the
area of ANNs and its their belief that, since there is no mapping
between input vectors and output vectors, an ANN can be an overkill
(not to mention it being ambiguous to describe the respective vectors).

Any pointers about what kind of statitstical distribution tests I can
run on the data to determine the distribution characteristics, and then
how I can try the estimation will be very helpful. Also will be
helpful, any papers which deal with this kind of data.

Thank You
Gangadhar

Google