Federal University of Ceara Department of Teleinformatics Engineering Elasticity in Cloud Computing Flávio R. C. Sousa [email protected] @flaviosousa www.lia.ufc.br/~flavio Workshop – LSBD – 2015 

Elasticity 2 Herbst et al. Elasticity in Cloud Computing: What It Is, and What It Is Not. In: ICAC. 2013. 

Elasticity “Capacity at runtime by adding and removing resources without service interruption in order to handle the workload variation” COUTINHO et al. Elasticity in cloud computing: a survey. In: Ann. Telecommun. 2014. Demand Capacity Time Resources 

Elasticity is the degree to which a system is able to adapt to workload changes by provisioning and de-provisioning resources in an autonomic manner, such that at each point in time the available resources match the current demand as closely as possible. 4 Herbst et al. Elasticity in Cloud Computing: What It Is, and What It Is Not. In: ICAC. 2013. 

Elasticity 5 Herbst et al. Elasticity in Cloud Computing: What It Is, and What It Is Not. In: ICAC. 2013. 

Elasticity advantages Customer • Ability to adjust resource consumption according to unpredicted demand. • Ability to free resources when they are not needed allowing to reduce costs. Provider • Allows VM consolidation and reduced server usage. • Keep SLAs/cluster occupancy ratio with the minimum resources possible. • Increase profit. 6 Beernaert et al. Automatic Elasticity in OpenStack. In: SDMCMM. 2013. 

Elasticity currently Manual (e.g., Amazon) Coarse-grained (usually at the VM level) No wide applicability (vendor lock-ins) Elasticity modeled as single-dimension property • Resource, cost and quality • Customization 7 CELAR Project 

Elasticity 8 Galante et al. A Survey on Cloud Computing Elasticity In: UCC. 2012. 

COUTINHO et al. Elasticity in cloud computing: a survey. In: Ann. Telecommun. 2014. Elasticity 

Elasticity 10 Turowski et al. Vertical Scaling Capability of OpenStack. In: ICSOC. 2014. 

Elasticity 11 Turowski et al. Vertical Scaling Capability of OpenStack. In: ICSOC. 2014. 

Elasticity Group Metris Allocation Computing Resource Allocation Meter (CRAM), Resource Allocation Capacity Available Supply, Capacity, System Capacity Meter (SCM), System Effective Capacity Meter (SEC) Cost Cost/ Performance rate, Cost Bandwidth, Migration Cost, Total Cost of Deployment QoS % Violations, SLA, System Performance Meter (SPM) Resource Utilization % Utilization, Cloud Stress, Computing Resource Utilization Meter (CRUM) Scalability Effective Scalable Range (ESR), Effective System Scalability (ESS), Scale-up Time Boot, Creation, Deletion, Resource Deallocation, Startup COUTINHO et al. Elasticity in cloud computing: a survey. In: Ann. Telecommun. 2014. 

Elasticity COUTINHO et al. Elasticity in cloud computing: a survey. In: Ann. Telecommun. 2014. 

14 Elasticity SOUSA, F. R. C. ; MACHADO, J. C. A Towards Elastic Multi-Tenant Database Replication with Quality of Service. In: UCC, 2012. 

OPENSTACK OPENSTACK OPENSTACK OPENSTACK ELASTICITY ELASTICITY ELASTICITY ELASTICITY 

OpenStack 16 

OpenStack 17 

OpenStack Ceilometer Collect from OpenStack components Transform metering data if necessary Publish meters to any destination (including Ceilometer itself) Store received meters Aggregate samples via a REST API 18 

OpenStack Heat 19 

OpenStack Heat 20 

OpenStack Heat 21 

OpenStack Heat 22 

OpenStack Heat 23 

OpenStack: Elasticity Vertical 24 Turowski et al. Vertical Scaling Capability of OpenStack. In: ICSOC. 2014. 

OpenStack: Elasticity Vertical 25 Turowski et al. Vertical Scaling Capability of OpenStack. In: ICSOC. 2014. 

OpenStack: Elasticity Vertical 26 Turowski et al. Vertical Scaling Capability of OpenStack. In: ICSOC. 2014. 

OpenStack: Elasticity Horizontal (Replication) Depending on when the updates are propagated • Eager • Lazy Depending on where the updates can take place • Centralized • Distributed Eager Lazy Centralized Distributed 

Elasticity: Challenges Fully automated • No manual setup Fine-grained • Not strictly at VM-level Real-time • Adaptive to changes in load, infrastructure and objective function • Prediction Elasticity to guarantee QoS • SLA to cloud storage 28 

Thank you! Thank you! Thank you! Thank you! Flávio R. C. Sousa [email protected] @flaviosousa www.lia.ufc.br/~flavio