-- I have data set of cpu and memory utility of 100 machines present in the network for the past 3 months and values are collected every 5 mins.
-- I want to define a neural network model for finding if the machine is compute intensive or not.
-- CPU is measured in cpu % usage (as we see in task manager) : ranging from 0 to 100 ... 100 is when 100% cpu is used.
Memory range in MB 500- 700 MB
-- I have converted the cpu and memory usage in the range of [0-1]
by using the fromula :
new val = (val - min) / max - min
where max , min and avg is the cpu or memory usage value.
-- I then mapped cpu and memory usage for all timestamps.
-- A term Work load is calculated as
W = cpu x mem
-- Now I want to design a neural network so tht i can define machine as compute intensive or not.
or
define at every timestamp of the machine as compute intensive task or not where again if the no. of tasks is being compute intensive is higher will now define if m/c as compute intensive.
Yk ,
I would suggest that you restate the problem in terms of tasks as opposed to machine.
Second, you have defined memory CPU ( I would assign a weight to memory CPU usage) but you have not defined I/O, which typically can be network card load or HD data transfers ( I would also assign a weight to these parameters).
Since I am assuming that you want to do this prediction to balance the load on your machines, I would also do an analysis of what is the time it takes for the typical tasks to complete so that you can have a better estimate of what time frame you can use (1 reading two readings or higher).
Hope this helps
Hi Yk Sr,
You can try Deep learning it can be a solution for your problem or you can use other technique, please take a look at links in topic.
-Neural Networks: Machine Learning Inspired by the Brain - Upwork
https://www.upwork.com › Hiring Headquarters Home › Data
-A Parallel Adaboost-Backpropagation Neural Network for Massive ...
https://www.nature.com › scientific reports › articles
-Adaptive Resource Utilization Prediction System for Infrastructure as a ...
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5547731/
-Big Data technologies: A survey - ScienceDirect
https://www.sciencedirect.com/science/article/.../S13191578173000...
-Energy Forecasting for Event Venues: Big Data and Prediction ...
https://www.sciencedirect.com/science/article/.../S03787788153044...
-Scaling Distributed Machine Learning with System and Algorithm Co ...
https://www.cs.cmu.edu/~muli/file/mu-thesis.pdf
-Multi-tenant GPU Clusters for Deep Learning Workloads - Microsoft
https://www.microsoft.com/.../Multi-tenant_GPU_Clusters-5af0bec...
Best regards
You must specify the number of neurons for the input and output also specify the number of layers.
Seems like a binary classification task , use a SVM as opposed to neural network , will be faster
- Badges
- Science topic