A Virtual Machine Placement Algorithm for Energy Efficient Cloud Resource Reservation
- 1. Topic : AVirtual Machine Placement Algorithm for Energy Efficient Cloud Resource Reservation CS399 (Seminar) Name : Suvom Das Roll No : 1501CS44 Stage : 2
- 2. Stage 1 - (Summary) Virtual Machine - Definition and advantages Power management techniques : specially consolidation Consolidation : 1)VM Placement and 2)VM Migration Different types of VM Placement Algorithms Graphical comparison between all the algorithms
- 3. Abstract - (Stage 2) What we are going to discuss in this seminar : In this seminar, we are going to discuss a new graph colouring model for advance resource reservation with minimum energy consumption in heterogeneous IaaS cloud data centres. We will start with an exact integer linear programming (ILP) formulation which generalises the graph colouring problem mathematically and follow with a Energy Efficient Graph Pre colouring (EEGP) heuristic to address the scalability and to reduce convergence times. The results of performance evaluation and comparisons of EEGP with the exact algorithm will demonstrate the efficiency of EEGP for the energy efficient advance resource reservation problem. We will see the efficiency of the EEGP algorithm by comparing it with the exact integer linear programming solution
- 4. Terminology PPW - (Performance Per Watt) : The term performance-per- watt is a measure of the energy efficiency of a computer architecture or a computer hardware. VRU - (Virtual Resource Unit) : Virtual units provided by the physical servers as per user request RRU - (Requested Resource Unit) : An RRU is an abstract representation of a VRU at request level.
- 6. Problem Formulation In order to solve the energy efficient advance cloud resource reservation problem using graph colouring and the PPW metric, we start from the representation depicted in Figure 1 of user requests and cloud data centre resources. Each request k, represented by Rk, consists of a user demand for nk virtual machines (VMs from V M1 to V Mnk ). We translate each VM request into requested resource units (RRUs) that have a one to one mapping with the VRUs provided by the infrastructure. The idea is to build a graph G where vertices represent requested resource units (RRUs) that will be associated to VRUs and links represent time overlap between RRUs. With this representation the advance resource reservation consists in finding an optimal mapping between the requested resources, expressed in RRUs, and the infrastructure servers whose capacity is measured in VRUs.
- 8. Graph colouring for Energy Efficient Resource Reservation The proposed energy efficient graph colouring is derived from an undirected dynamic graph G = (V,E). G is dynamically constructed over time and updated after request arrivals and departures. Vertex set V represents RRUs belonging to all requested VMs and E represents the set of all edges. When a new request arrives (request Rk), new nodes and edges appear on the graph. The resulting global graph (G = G ∪ Gk) is a conjunction of G and Gk (comprising the set of VM nodes required on the request Rk and their edges). When a reservation ends, due to a departure or end of service, Gk will be retrieved and deleted from the global graph G.
- 10. GRAPH COLOURING ALGORITHMS FOR ENERGY EFFICIENT CLOUD RESOURCE RESERVATION Since the energy efficient VM reservation problem in cloud data centres is known to be NP-hard, an exact (in our case an ILP-based algorithm) will find the optimal solutions in acceptable convergence times only for small graphs or problem sizes. The exact algorithm is useful to check if the performance of the EEGP algorithm is close to optimal in terms of number of used colours and energy efficiency and to assess the performance improvement in convergence time.
- 11. Energy efficient graph per colouring heuristic The proposed EEGP algorithm assigns gradually colours to not yet coloured vertices (RRUs). The algorithm uses the steps specified below to achieve colouring which is equivalent to assigning a VRU to each RRU in the set. The EEGP algorithm uses the following steps to find a solution: Step 1 : Find the colour cluster Cj with the highest PPW and with free VRUs. The first step of the EEGP algorithm is handled by the function Find-Color- Cluster(C ,VMi )
- 12. Energy efficient graph per colouring heuristic Step 2 : Determine the neighbouring RRUs (or graph vertices) directly connected to V Mi RRUs. This step in the EEGP algorithm is handled by the function List-of-Connected- Nodes(V Mi) that constructs the list LV Mi of RRUs connected to V Mi. In Figure 4, LV M4 = {v2, v3, v4}. Step 3 : Construct the list of colours ∈ Cj that are not assigned to VMi neighbouring RRUs. This step uses the function List-of-Unused-colors(Cj ,L VMi ) to construct the list colj,V Mi .
- 13. Energy efficient graph per colouring heuristic Step 4 : Finally, the algorithm can assign to each RRU ∈ V Mi a different colour from the list col j ,V Mi
- 14. Evaluation Environment For Dataset : Requests are generated by poisson arrivals at a rate of 10 requests per second. Demand duration ( or time during it each V Mi is reserved (bi − ai) ) is uniformly distributed between 200s and 1800s. The number of VMs per request is uniformly distributed between 1 and 10. The number of VM nodes per VM is uniformly distributed between 1 and 20 ( e.g EC2 instance types, namely small, large, xlarge, high-cpu-medium and high-cpu-xlarge have respectively 1, 4, 8, 5 and 20 EC2 compute units (ECUs)). We will compare the result from EEGP to Haizea advance reservation (AR) algorithm too.
- 15. Performance Result Average performance per watt vs graph size (no of request)
- 16. Performance Result Number of used servers vs graph size (no of request)
- 17. Performance Result Convergence time vs graph size (no of request)
- 18. Performance Result Convergence time vs graph size (no of request)
- 19. Performance Result Chromatic number vs graph size (no of request) Chromatic number : Number of used colours
- 20. Improvements If we see the above graph formation technique used for EEGP : Graph will be more complex if the request demand for a larger instance. So that if a request of large-VM (consist of 20 RRUs) arrives then the graph size will increase by 20 but 20C2 new edges will increase the complexity further.
- 21. Further Work : (Problems) If we see the graph (Convergence time vs Graph size), we can see that the graph is increasing exponentially. So for increasing number of request it will take more and more time for placement. Solution : A graph multi colouring problem which will assign VM requests as a instances.
- 23. References A Virtual Machine Placement Algorithm for Balanced Resource Utilisation in Cloud Data Centres Exact and Heuristic Graph Colouring for Energy Efficient Advance Cloud Resource Reservation
- 24. Thank You