Modeling and Optimization of Resource Allocation in Cloud [PhD Thesis Progress 4]
1 like488 views
This document outlines a PhD thesis by Atakan Aral focused on optimized resource mapping in distributed cloud systems, specifically discussing the distributed replica placement for edge computing. It details the thesis structure, including evaluation methods using real-world datasets, and presents algorithms aimed at improving latency and cost-efficiency in data replication. The work's significance lies in its approach to resource allocation and replica management in cloud environments to enhance performance and availability.
![Introduction
Distributed Replica Placement for Edge Computing
Evaluation and Discussion
Overview of the Thesis Study
Summary of the Term
Time Plan
Evaluation
Implemented and Simulated in CloudSim
Real workload traces are used
The CAIDA Anonymized Internet Traces 2015 Dataset [1M requests]
1998 FIFA World Cup Web Site Requests Dataset [880K requests]
Baselines: Centralized storage, Caching
Parameters: Quantum length, level of expansion
Criteria: Data access latency, data storage and transfer cost, network
overhead, false positive hit rate
Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems](https://image.slidesharecdn.com/progress4-160916170521/85/Modeling-and-Optimization-of-Resource-Allocation-in-Cloud-PhD-Thesis-Progress-4-5-320.jpg)
![Introduction
Distributed Replica Placement for Edge Computing
Evaluation and Discussion
Overview of the Thesis Study
Summary of the Term
Time Plan
Journal Publication
Network-Aware Embedding of Virtual Machine Clusters onto Federated
Cloud Infrastructure
The Journal of Systems and Software, Elsevier [Impact Factor: 1.767]
Revised on 04.06.2016, accepted for publication on 06.07.2016
Available online with DOI: 10.1016/j.jss.2016.07.007
Revision includes extended related work, a use case scenario, and evaluation
of runtime performance.
Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems](https://image.slidesharecdn.com/progress4-160916170521/85/Modeling-and-Optimization-of-Resource-Allocation-in-Cloud-PhD-Thesis-Progress-4-6-320.jpg)
![Introduction
Distributed Replica Placement for Edge Computing
Evaluation and Discussion
Overview of the Thesis Study
Relative to Centralized Solution
[0.012, 2000]: 28.04% latency decrease 22.12% cost increase
[0.012, 4000]: 32.99% latency decrease 33.10% cost increase
Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems](https://image.slidesharecdn.com/progress4-160916170521/85/Modeling-and-Optimization-of-Resource-Allocation-in-Cloud-PhD-Thesis-Progress-4-28-320.jpg)
Modeling and Optimization of Resource Allocation in Cloud [PhD Thesis Progress 4]
- 1. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Optimized Resource Mapping in Distributed Cloud Systems PhD Thesis Progress – Fourth Report Atakan Aral Thesis Advisor: Asst. Prof. Dr. Tolga Ovatman Istanbul Technical University – Department of Computer Engineering July 12, 2016 Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 2. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 3. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Summary of the Term Time Plan Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 4. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Summary of the Term Time Plan Distributed Replica Placement (DRP) Algorithm A fully distributed algorithm for data replication in Edge Computing Systems that is based on Facility Location Problem in Operation Research that addresses the trade-off between price (storage and bandwidth) and performance (latency) that is dynamic, online, and incremental. that works with the limited knowledge of the system. Benefits include low latency access to centralized data better price–performance ratio than caching reduced network overhead Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 5. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Summary of the Term Time Plan Evaluation Implemented and Simulated in CloudSim Real workload traces are used The CAIDA Anonymized Internet Traces 2015 Dataset [1M requests] 1998 FIFA World Cup Web Site Requests Dataset [880K requests] Baselines: Centralized storage, Caching Parameters: Quantum length, level of expansion Criteria: Data access latency, data storage and transfer cost, network overhead, false positive hit rate Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 6. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Summary of the Term Time Plan Journal Publication Network-Aware Embedding of Virtual Machine Clusters onto Federated Cloud Infrastructure The Journal of Systems and Software, Elsevier [Impact Factor: 1.767] Revised on 04.06.2016, accepted for publication on 06.07.2016 Available online with DOI: 10.1016/j.jss.2016.07.007 Revision includes extended related work, a use case scenario, and evaluation of runtime performance. Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 7. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Summary of the Term Time Plan Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 8. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Summary of the Term Time Plan Gantt Chart 2016 1 2 3 4 5 6 Implementation Evaluation Literature Review Documentation TBM Revision TBM Submission Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 9. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 10. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Distributed Replica Placement Motivation Edge Computing Pushing the frontier of computing applications, data, and services away from centralized nodes to the logical extremes of a network (e.g. mobile devices, sensors, nano data centers, routers, modems, . . . ) Provides low-latency access to computing resources for code offloading. However, many applications still need to access data that is stored centrally Due to the limited storage capacity of the edge entities, economic constraints, availability for offline analysis, simpler maintenance and concurrency control Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 11. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Distributed Replica Placement Data Replication A frequently used technique to improve availability, fault-tolerance, security, and/or access latency. 1 Which data to replicate? 2 When to replicate? 3 Where to place the replica? 4 How to direct requests to replicas? 5 How to keep replicas consistent? 6 How to prefetch data to exploit spatial locality? Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 12. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 13. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Facility Location Heuristic Solution Idea Main Idea Create replicas of the data object on locations where access paths for that object frequently pass through so that future requests from multiple locations can be served from a single replica. We assume temporal and geographical locality of reference. We also assume that geographical distance is correlated with latency. Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 14. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Facility Location Heuristic Temporal and Geographical Locality Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 15. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Facility Location Heuristic Temporal and Geographical Locality Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 16. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Facility Location Heuristic Facility Location Problem Finding a placement of facilities in order to serve the demands of geographically distributed customers with minimum cost (transportation and facility building) minimize j fj · Yj + i j hi · dij · Xij Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 17. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Facility Location Heuristic minimize j fj · Yj + i j hi · dij · Xij Facility opening cost (fj) ∝ storage cost for the replica in the next quantum. fj = unit_pricej · replica_size · quantum Demand (hi) ∝ number of requests received in the previous quantum hi = num_requests · replica_size Distance (dij) ∝ latency between a VM and a replica dij = latencyij · λ Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 18. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Facility Location Heuristic minimize j unit_pricej · quantum + i j num_requests · latencyij · λ λ is the unit conversion factor. It represents the expendable unit cost in exchange for a unit decrease in latency per unit demand. It can be tuned for different service level objectives in terms of latency. Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 19. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Facility Location Heuristic minimize j unit_pricej · quantum + i j num_requests · latencyij · λ In every quantum, locally evaluate the objective function at the central storage to Create replicas in the neighbours. In every quantum, locally evaluate the objective function at each replica location for one of the following operations: Remove the replica at that location Duplicate the replica to one of the neighbour locations Migrate the replica to one of the neighbour locations Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 20. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 21. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Replica Discovery How to notify an edge entity when a closer replica is created? Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 22. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Facility Location Heuristic Replica Discovery Replica Discovery 1 When a DC decides to deploy a replica in one of its neighbours, it notifies previous requesters. 2 When a new replica is created in a location, its neighbours are notified. 3 When any message is received, DC calculates and stores the latency to the source of the message 4 When a VM requires a data object, list of previous notifications is searched. 5 If there are multiple candidates, the decision is made based on known latencies. 6 Otherwise, the data is requested from the main storage. Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 23. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 24. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Lambda Parameter Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 25. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Lambda Parameter Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 26. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Quantum Duration Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 27. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Quantum Duration Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 28. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Relative to Centralized Solution [0.012, 2000]: 28.04% latency decrease 22.12% cost increase [0.012, 4000]: 32.99% latency decrease 33.10% cost increase Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 29. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 30. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Theme "Optimized Resource Mapping in Distributed Cloud Systems" Mapping virtual entities to physical cloud resources at hand Resources are distributed and networked Decreasing the network latency and optimizing bandwidth utilization so as to improve the performance of cloud services Also considering the resource costs to improve price-performance ratio Employing approximation algorithms and/or heuristics to solve NP-complete problems dynamically Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 31. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Use Case Scenario A Cloud Broker or Mediator Has clients who wish to execute their data-intensive tasks with high performance and availability. Clients demands SLOs in terms of average global latency or total execution time. Hires cloud resources from a IaaS provider to fulfill client needs. Has to minimize its operating costs but also satisfy SLOs Input: VM Cluster to be executed, required data objects, SLOs Output: mapping between VMs and compute servers (usually static), mapping between data objects and storage servers (usually dynamic) Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 32. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 33. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Part I: Virtual Machine Mapping Topology based Mapping (TBM) Algorithm Map VM Clusters onto the federated cloud infrastructure based on their topology. Aral, A. and Ovatman, T. (2014). Improving resource utilization in cloud environments using application placement heuristics. In Proceedings of the 4th International Conference on Cloud Computing and Services Science (CLOSER), pages 527–534. Aral, A. and Ovatman, T. (2015). Subgraph matching for resource allocation in the federated cloud environment. In Proceedings of 8th IEEE International Conference on Cloud Computing (IEEE CLOUD), pages 1033–1036. Aral, A. and Ovatman, T. (2016). Network-aware embedding of virtual machine clusters onto federated cloud infrastructure. The Journal of Systems and Software. http://dx.doi.org/10.1016/j.jss.2016.07.007 Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 34. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Outline 1 Introduction Summary of the Term Time Plan 2 Distributed Replica Placement for Edge Computing Facility Location Heuristic Replica Discovery 3 Evaluation and Discussion 4 Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 35. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Part II: Data Replica Mapping Distributed Replica Placement (DRP) Algorithm Map replicas to the locations where its access paths frequently pass through. Aral, A. (2016). Network-aware resource allocation in distributed clouds. In Doctoral Symposium of the IEEE International Conference on Cloud Engineering (IC2E). Aral, A. and Ovatman, T. (In preparation). Latency- and cost-aware distributed replica placement for edge computing. Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems
- 36. Introduction Distributed Replica Placement for Edge Computing Evaluation and Discussion Overview of the Thesis Study Virtual Machine Mapping Data Replica Mapping Thank you for your time. Atakan Aral Optimized Resource Mapping in Distributed Cloud Systems