Outline Research Problem Research Problem Challenges Approaches - PowerPoint PPT Presentation

8/7/2012 Outline  Research Problem Research Problem  Challenges  Approaches & Gaps PHD Dissertation Proposal Defense  Research Goals  Research Questions & Experiments  Research Questions & Experiments Wes J. Lloyd  Research Contributions August 7, 2012  Preliminary Results Colorado State University, Fort Collins, Colorado USA August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense 2 Infrastructure as a Service (IaaS) Traditional Application Deployment Cloud Computing App pp  Server partitioning of multi ‐ core servers  Server partitioning of multi ‐ core servers Data Data Business Business Logging Logging Server  Hardware virtualization Spatial DB  Service isolation Services  Resource elasticity Tracking Apache rDBMS DB Tomcat Services Services Object DODB / Single Server Store Services NOSQL August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Problem 3 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Problem 4 1

8/7/2012 Problem Statement  Autonomic deployment of multi ‐ tier  Autonomic deployment of multi ‐ tier applications to IaaS clouds  Component composition  Collocation and interference of components  Scaling infrastructure to meet demand  Scaling infrastructure to meet demand August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Problem 5 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Problem 6 Application Component Composition Application Components Virtual Machine (VM) Virtual Machine (VM) Images App Server App Server rDBMS write rDBMS r/o Component File Server File Server Deployment Log Server Log Server Image 1 Image 2 Load Balancer rDBMS r/o rDBMS write Load Balancer Load Balancer Dist. cache . . . Image n Application “Stack” PERFORMANCE August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense 7 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Problems & Challenges 8 2

8/7/2012 Provisioning Variation Bell’s Number Request(s) to n = number VMs Share PM launch VMs of CPU / Disk / Network n k application application VM components VM deployments 4 15 VM VM VM VM VM VM Ambiguous config 1 config 2 5 52 Mapping VM VM M odel M D M VM VM D F 1 VM : 1..n components 6 203 D atabase L L F Physical Host Physical Host Physical Host VM Component 7 877 F ile Server Deployment VM VM VM VM config n VM L og Server 8 4,140 M L D VM 9 21,147 Application F . . . “Stack” Physical Host Physical Host Physical Host n . . . k= # of VMs Reserve # of Configurations PERFORMANCE possible PM Memory Blocks configs Problems & Challenges August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Problems & Challenges 9 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense 10 Infrastructure Management Service Requests • Scale Services • Tune Application • Tune Application Load Balancer Parameters • Tune Virtualization Application Servers Parameters Load Balancer distributed cache noSQL data stores rDBMS August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Problems & Challenges 11 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense 12 3

8/7/2012 Virtual Machine (VM) Placement Related Work as “Bin Packing Problem”  Bins= physical machines (PMs)  Multivariate performance models Bins physical machines (PMs) Multivariate performance models  Items= virtual machines (VMs)  Regression models  Dimensions  Machine learning  CPU time  Feedback loop control  VM RAM, hard disk size, # cores  Hybrid approaches  Disk read/write throughput  Disk read/write throughput  Formal approaches  Network read/write throughput NP-Hard  Integer linear programming  PM capacities vary dynamically  Case based reasoning  VM resource utilization varies August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Approaches & Gaps 13 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Approaches & Gaps 14 Gaps in Related Work Why Gaps Exist  Existing approaches do not consider  Public clouds  Public clouds  VM image composition  VM image composition  Complementary component placements  Research is cost prohibitive  Interference among components  Users concerned with performance not in control  Minimization of resources (# VMs)  Private clouds: systems still evolving  Load balancing of physical resources  Performance models (large problem space) Performance models (large problem space)  Performance models ignore  Performance models ignore  Disk I/O  Virtualization misunderstood or overlooked  Network I/O  VM and component location August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Approaches & Gaps 15 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Approaches & Gaps 16 4

8/7/2012 Research Goals RG1: Support VM component composition RG2: Support virtual infrastructure management  Determine and execute VM placement  Scale infrastructure for application demand August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense 17 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Goals 18 Performance Objectives  Primary: Maximize application throughput  Primary: Maximize application throughput  Secondary: Minimize resource cost (# of VMs)  Minimize modeling time  Support high responsiveness to change in application demand application demand August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Goals 19 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense 20 5

8/7/2012 Methodology Evaluation Component Infrastructure Composition Management  CSIP: USDA ‐ NRCS platform for model services p RQ3  Models as multi ‐ tier application surrogates RQ1  RUSLE2 – Soil erosion model  WEPS – Wind Erosion Prediction System RQ4  Hydrology models: SWAT, AgES  Other models: STIR, SCI…  Other models: STIR SCI  Eucalyptus IaaS cloud(s) RQ2 RQ5  Amazon EC2 compatible  XEN & KVM hypervisors August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Methodology 21 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Methodology 22 RQ1: Which independent variables best help RQ1: Which independent variables best help model application performance (throughput) model application performance (throughput) to guide autonomic component composition? to guide autonomic component composition? Methodology  Total (all VMs) resource utilization  Total (all VMs) resource utilization  Total (all VMs) resource utilization  Total (all VMs) resource utilization Exploratory performance modeling Exploratory performance modeling  CPU time, disk I/O, network I/O, …  CPU time, disk I/O, network I/O, … • Investigate independent variables  Individual VM and PM resource utilization  Individual VM/PM resource utilization • Investigate modeling techniques  Component and VM location  Component / VM location • Multiple linear regression (MLR) • Artificial neural networks (ANNs) • Artificial neural networks (ANNs)  VM Configuration: number of cores, RAM,  VM Configuration: number of cores RAM  VM Configuration: number of cores, RAM,  VM Configuration: number of cores RAM • Others hypervisor type (KVM, XEN...) hypervisor type (KVM, XEN...) August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Methodology 23 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Methodology 24 6

8/7/2012 RQ2: Can component resource classifications RQ2: Can component resource classifications and behavioral rules predict performance of and behavioral rules predict performance of component compositions? component compositions? Methodology  Support simplification of the search space  Support simplification of the search space  Support simplification of the search space  Support simplification of the search space Investigate autonomic component composition Investigate autonomic component composition approach(es)  Support applications with large # of  Support applications with large # of n k n k components components 4 15 4 15 • Performance modeling  Bell’s number 5 52  e.g. Bell’s number: 5 52 • Heuristics to classify 6 203 6 203 • Component resource utilization • Component resource utilization 7 877 7 877 • Component dependencies 8 4,140 8 4,140 9 21,147 9 21,147 n . . . n . . . August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Methodology 25 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Methodology 26 RQ3: Does performance of Evaluation Metrics: component compositions change Component Composition when scaled up?  Composition performance C i i f  Single provisioned application VMs   Average throughput of configurations Multiply provisioned application VMs  Resource packing density  Investigate collocation of new VMs  # components/# VMs for compositions  Intelligent vs. ad ‐ hoc placement  Derivation speed D i ti d  Load balance physical resources  Average wall clock time to produce compositions August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Methodology 27 August 7, 2012 Wes J. Lloyd PHD Dissertation Proposal Defense Research Questions & Experiments 28 7