cloud-Application-Presentation-Virtual-Machine.pptx

Cloud Computing
Machine

1. An Optimum Compute Resource Consolidation Framework for Cloud Data Center Submitted By: Sheetal Garg Enroll No: 901903011 Supervisors : Dr. Raman Singh (Assistant Professor) Dr. Rohit Ahuja (Assistant Professor) Dr. Ivan Perl (Associate Professor, ITMO University, Russia)

2. Name SHEETAL GARG Registration Number 901903011 Date of Admission 31st July 2019 Date of URB/IRB 02nd June 2021 Title An Optimum Compute Resource Consolidation Framework for Cloud Data Center Name of the Supervisor(s) Dr. Raman Singh, D r. Rohit Ahuja, Dr. Ivan Perl Percentage of completion of Ph.D. work as mentioned in the previous progress monitoring form 1 st Progress Monitoring

3. Introduction What is Cloud Computing? Cloud computing is the use of computing resources that are delivered as a service over a network [7]. . Figure 1: Cloud Computing Architecture

4. Challenges in Cloud Data Center Cloud service providers deals with several challenges in order to provide cloud services to users: Energy consumption management Effective resource utilization Detect Hotspot and Cold spot Load balancing Response time Virtual machine migration

5. Research Objectives To study and analyze existing virtual machine placement, selection, consolidation and resources prediction techniques. To propose workload prediction model for resources utilization of cloud data center . To propose optimum compute resources consolidation framework (using virtual machine placement and selection) for energy-efficient cloud data centers. To validate and evaluate the performance of the proposed techniques of workload prediction and virtual machines consolidation.

6. Literature Review for Workload Prediction Author Name Year Dataset Technique Remarks Salam Ismaeel et.al. [1] 2015 Real Google trace K mean and Extreme Learning Machine (ELM) Aim : Developed a prediction model for estimate the future virtual machine requests in a data center . Technique used: combining a k-means clustering algorithm and an Extreme Learning Machine (ELM ). Result : This model able to estimate requests in all clusters with minimum error compared to linear regression. Milad Seddigh et.al. [2] 2015 Random generated values Virtual machine dynamic prediction scheduling via ant colony optimization (VMDPS-ACO) Aim : Predict the memory usage on the PMs to prevent overloading of PMs and migration of virtual machines from a PM to another PM. Result : This algorithm can save 52% physical resources compared to the worst version of First Fit Decrease ( FFD) and 28% lower resource wastage compared to the best version of FFD in the homogeneous mode.

7. Author Name Year Dataset Technique Remarks Jitendra Kumar et.al. [3] 2017 NASA-HTTP, Calgary-HTTP, Saskatchewan-HTTP Long Short term Memory (LSTM) Aim: Predict the workload of cloud datacenter. Results : Minimum error achieved for D1, D2, and D3 are 0.00479, 0.00342 and 0.00317 respectively. The proposed model results were compared with back propagation and outcomes of experiments clearly convey that the LSTM-RNN based forecasting model outperforms. Jing Chen et.al. [4] 2018 Random values EEMD-ARIMA (Ensemble empirical mode decomposition ) Aim : Predict the future resource demands accurately to support resource provision in advance. Result : The results showed that EEMD-ARIMA method is more effective than ARIMA model in terms of RMSE. ARIMA model achieved RMSE of 2.45 and EEMD-ARIMA RMSE of 1.97. Yanxin Liu et.al. [5] 2019 Cluster Google trace Moving average , ARIMA, SVR and Back P ropagation Aim: Analyzes the performance of four different prediction algorithms : Moving average, ARIMA, SVR and back propagation. Result : If have enough training time then use SVR. Its prediction time is also less. If does not have much time to train and data amount is also less than use ARIMA model. BPNN used where pattern cannot be analyzed by human.

8. Author Name Year Dataset Technique Remarks Sun-Yuan Hsieh et.al. [6] 2020 PlantLab data Gray- Markov Model Aim: Predict short term future CPU utilization by using historical data . Result : The proposed approach reduces energy consumption by an average of 42.7%, 38.1%, 39%, and 33.1% compared with Static Threshold, Inter-quartile Range, Mean Absolute Deviation, and Local Regression respectively. It also minimize unnecessary VM migrations and the number of active hosts to economize. Nguyen Trung Hieu et. al. [7] 2020 Google trace Virtual Machine Consolidation With Multiple Usage Prediction (VMCUP-M) Aim: Proposed algorithm is executed during the virtual machine consolidation process to estimate the long-term utilization of multiple resource types based on the local history of the considered servers. Result : The proposed algorithm effectively reduces the number of active servers, migrations, power state changes and the energy consumption of the servers. VMCUP-M reduced about 78.26 percent of the migrations. Qazi Zia Ullah et.al. [8] 2021 Bitbrain Cartesian genetic programming neural network (CGPNN) Aim: Cartesian genetic programming (CGP) based parallel neuro -evolutionary prediction model to solve the CPU usage prediction problem. Result: This model achieved 97% prediction accuracy, which will lead to correct scaling decisions in predictive scaling mechanisms of cloud servers.

9. Work done till now Completed coursework from August - December, 2019 with CGPA 9.0/10. Review the literature:- Prediction of Virtual machine resources utilization for Cloud Data Center and Virtual machine consolidation. Download the Google cluster usage trace version 2.1 dataset by Google cloud SDK tool. Convert the dataset into time Series of 5 min interval. Exact the relevant research related features i.e. CPU, Memory and Hard disk utilization of machine. Implementing the Univariate Time Series Long Short term Memory (LSTM) model.

10. Results 45 min: MSE- 0.0464 5 min: MSE- 0.0435 60 min: MSE- 0.0466 15 min: MSE- 0.0442 30 min: MSE- 0.0466 Figure 2: LSTM on 12 days data of Physical Machine 5

11. Planning of work for next semester Analysis of LSTM model and compare the result with traditional models. Communicate research paper based on results observed.

12. References 1. Ismaeel , Salam, and Ali Miri . "Using ELM techniques to predict data centre VM requests." In 2015 IEEE 2nd International Conference on Cyber Security and Cloud Computing, pp. 80-86. IEEE, 2015. 2. Seddigh , Milad , Hassan Taheri , and Saeed Sharifian . "Dynamic prediction scheduling for virtual machine placement via ant colony optimization." In 2015 Signal Processing and Intelligent Systems Conference (SPIS), pp. 104-108. IEEE, 2015. 3. Kumar, Jitendra , Rimsha Goomer , and Ashutosh Kumar Singh. "Long short term memory recurrent neural network (LSTM-RNN) based workload forecasting model for cloud datacenters." Procedia Computer Science 125 (2018): 676-682. 4. Chen, Jing, and Yinglong Wang. "A resource demand prediction method based on EEMD in cloud computing." Procedia Computer Science 131 (2018): 116-123. 5. Liu, Yanxin , Jian Dong, Decheng Zuo , and Hongwei Liu. "Experimental Analysis and Comparison of Load Prediction Algorithms in Cloud Data Center." In 2019 IEEE 19th International Conference on Software Quality, Reliability and Security (QRS), pp. 197-203. IEEE, 2019. 6. Iqbal , Waheed , Josep Lluis Berral , and David Carrera . "Adaptive sliding windows for improved estimation of data center resource utilization." Future Generation Computer Systems 104 (2020): 212-224. 7. Hieu , Nguyen Trung , Mario Di Francesco, and Antti Ylä-Jääski . "Virtual machine consolidation with multiple usage prediction for energy-efficient cloud data centers." IEEE Transactions on Services Computing 13, no. 1 (2020): 186-199. 8. Ullah , Qazi Zia, Gul Muhammad Khan, Shahzad Hassan, Asif Iqbal , Farman Ullah , and Kyung Sup Kwak . "A Cartesian Genetic Programming Based Parallel Neuroevolutionary Model for Cloud Server’s CPU Usage Prediction." Electronics 10, no. 1 (2021): 67.

13. Thank You