SLAMMP Framework for Efficient Resource Monitoring and Prediction at an IaaS Cloud by Vivek Kumar Prasad

Guided by: Dr Madhuri Bhavsar With Synopsis and CD
15EXTPHDE145

ABSTRACT:
The Cloud Computing(CC) paradigm has transformed the information technology
horizon in past years and has emerged as an important computing utility. Government
bodies, industries and academia have given significant attention to the CC. Cloud has
become the backbone of the current economy by posing subscription-based facilities anywhere,
anytime resulting in a pay-as-you-go model. CC supports properties such as scalable
resources handling and elasticity through resource management. The management of
the resources is being handled through monitoring and prediction . The present challenge
in CC environment is to identify the possible violations in the SLA proactively. Also
reacting to this state by taking appropriate actions to avoid penalties and manage the
resources effectively. This research work addresses resource monitoring and prediction
mechanism to handle users’ demands in an efficient way in the CC environment through
the Service Level Agreement Management using Monitoring and Prediction (SLAMMP)
framework.
The framework integrates the concepts of Deep Learning (DL), Hidden Markov Model
(HMM), and Smart Contracts (SC); and is mapped to four-fold layers. First, the workload
generation has been implemented through Reinforcement Learning (RL); secondly,
the anti-patterns of the workloads were checked by using HMM, thirdly the SLAs has
been maintained using a smart contract, and fourth, is the utilization of resources has
been predicted using Long Short Term Memory (LSTM) approach. The SLAMMP framework
discussed here ensures timely monitoring and prediction of the cloud infrastructure,
which results in the analysis of the realistic (real-time) behavior of the IaaS cloud and
take precautionary actions for the management of cloud resources during peak time/high
demand. This mechanism is better for capacity planning, Admission control, and SLA
process management. The experiment shows that the proposed SLAMMP framework
effectively manages the cloud resources using monitoring and prediction methodologies.
The SLAMMP framework is evaluated against other techniques such as Artificial Neural
Network (ANN), Recurrent Neural Network (RNN), Random Forest (RF), Decision Tree
(DT), Support Vector Regression (SVR), Gated Recurrent Unit(GRU), and Autoregressive
integrated moving average (ARIMA) .
Various parameters such as CPU utilization, disk write throughput, disk read through put, memory usages, network received throughput, and network transmitted throughput
are used to validate the framework. The evaluation based on the performance metrics and
statistical t testing shows that the mentioned framework makes a substantial improvement
in resource management. The results are validated for both patterns and anti-patterns
based resource utilities. This also meets the SLA and also restricts the violations. Overall,
we conclude that based on the experimental results, the designed and implemented
SLAMMP framework works efficiently as compared to the other online machine learning
and deep learning techniques. The framework manages resources optimally while dealing
with the patterns and anti-patterns. This research work contributes towards an overall
performance and Quality of Service (QoS) enhancement for resource management in the
CC ecosystem.