ML-driven Classification Scheme for Dynamic Interference-aware Resource Scheduling in Cloud Infrastructure
This repository refers to a paper submitted to Journal of Systems Architecture.
https://www.journals.elsevier.com/journal-of-systems-architecture
Here, you will find all used codes and results from this work.
Authors: Vinícius Meyer, Dionatrã F. Kirchoff, Matheus L. da Silva and César A. F. De Rose
School of Technology, Pontifical Catholic University of Rio Grande do Sul (PUCRS) - Porto Alegre, Brazil
Reference: to appear
Classifier.R is the main class to run the interference-aware classification from a given profiled application. Few variables need to be filled, as follows:
app_tittle receives the profiled data from a given period in CSV format.
period refers to the segmentation period to be classified. For example: if it is set to 25, it means the total profiled data will be divided into four parts; 50% into two; and so on.
method set the method used outcome classification. In this paper, we only use "L" (level), which means that all classes of interference (CPU, Cache, Disk, Network, and Memory) will receive one of four defined levels (High, Moderate, Low, or Absent) using the K-means model. Other types of classifications were tested in preliminary experiments, such as: "C"(class) returns the major class of interference, this means that from a given profiled period, the most representative, or the interference class that suffers the most, will be exposed; or "P"(percent) that brings the percentage of each class of interference from a given period.
file result.pdf is the final classification result (plot).
training_dataset refers to data used to build the machine learning models (SVM and K-means).
source contains the profiled data from used applications (Bench4Q, LinkBench, and TPC-H).
accuracy presents the font codes to perform the accuracy evaluation phase.
If you have any question please do not hesitate to contact us through vinicius.meyer@edu.pucrs.br