mziccard / pyscale

A workload prediction library for Python

PyScale

A workload prediction library for Python. PyScale can be used to estimate future loads (in terms of traffic/CPU load/memory usage) of your web application/service to help you scale up/out proactively.

Load Prediction

PyScale uses Extreme Value Analysis (EVA) to predict the workload of your deployed services. To predict workload peaks, PySCale feds to Extreme Value Analysis a time serie representing a history of loads. This technique can be applied to different metrics as, for instance, traffic load, memory consumption or CPU usage. EVA fits to the provided data a continous probability distribution. From the corresponding survival function (1 - cumulative distribution function) it is possible to extract values that are only exceeded with arbitrarily low probabilities.

For instance, let's pick from the survival function computed via EVA a probability p = 0.001 and its corresponding load value x (where x indicates either traffic load, memory requirements or CPU usage). We now not only have a prediction of future load but we also know the probability that this prediction is exceeded. Scaling up/out our service so that is can handle a load of x means knowing that the probability of overloading the service is p = 0.001. That is, 99.9% of uptime.

Usage

To predict a load peak at a given probability you can instantiate an object of the class LoadPredictor as:

from pyscale import LoadPredictor

load_data = [ 4352, 4472, 3847, 4915, 4969, 4333, 4381, 4091, 4135, 4160,
              3534, 4598, 4086, 3788, 4038, 3396, 4118, 3822, 4333, 4034 ]
predictor = LoadPredictor(load_data)
load_peak = predict_load(0.001)

You can also produce a plot of the survival function by istantiating an object of the class PredictionPlotter:

from pyscale import LoadPredictor

load_data = [ 4352, 4472, 3847, 4915, 4969, 4333, 4381, 4091, 4135, 4160,
              3534, 4598, 4086, 3788, 4038, 3396, 4118, 3822, 4333, 4034 ]
predictor = LoadPredictor(load_data)
plotter   = PredictionPlotter(predictor)
plotter.xlabel('Requests')
plotter.ylabel('Probability')
plotter.plot("plot.png", 0.001)

Example

After cloning the repository you can move to the root of the project and run the example with:

python examples/example1.py

This example takes input data from two files in examples/data and produces in examples two plots representing load prediction made starting from the time series in the two data files.