Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Modelling and Managing Deployment Costs of Microservice-Based Cloud Applications

xLeitix
December 08, 2016

Modelling and Managing Deployment Costs of Microservice-Based Cloud Applications

Talk given at UCC'16

xLeitix

December 08, 2016
Tweet

More Decks by xLeitix

Other Decks in Science

Transcript

  1. software evolution & architecture lab University of Zurich, Switzerland Modelling

    and Managing Deployment Costs of Microservice-Based Cloud Applications Philipp Leitner, Jürgen Cito, Emanuel Stöckli @xLeitix
  2. Microservices FTW! Customer-Facing Web Interface /get order history /save /generate

    recommend /get customers /login orders /get /submit banking_ interface /process_ payment /landing frontend /order products /get
  3. Microservices FTW! Customer-Facing Web Interface /get order history /save /generate

    recommend /get customers /login orders /get /submit banking_ interface /process_ payment /landing frontend /order products /get
  4. Microservices FTW! Customer-Facing Web Interface /get order history /save /generate

    recommend /get customers /login orders /get /submit banking_ interface /process_ payment /landing frontend /order products /get
  5. Microservices FTW! Customer-Facing Web Interface /get order history /save /generate

    recommend /get customers /login orders /get /submit banking_ interface /process_ payment /landing frontend /order products /get
  6. Microservices FTW! Customer-Facing Web Interface /get order history /save /generate

    recommend /get customers /login orders /get /submit banking_ interface /process_ payment /landing frontend /order products /get
  7. A Not-so-Unrealistic Scenario Initial Version Updated Version frontend /landing recommend

    /generate products /get frontend /landing recommend /generate products /get orderhistory /get
  8. A Not-so-Unrealistic Scenario Initial Version Updated Version frontend /landing recommend

    /generate products /get frontend /landing recommend /generate products /get orderhistory /get
  9. A Not-so-Unrealistic Scenario Initial Version Updated Version frontend /landing recommend

    /generate products /get frontend /landing recommend /generate products /get orderhistory /get
  10. A Not-so-Unrealistic Scenario Initial Version Updated Version frontend /landing recommend

    /generate products /get frontend /landing recommend /generate products /get orderhistory /get
  11. Modelling Microservice Costs Basic intuitions: • Total costs of deployment

    = sum of all individual service costs • For each service: CostHat
  12. Modelling Microservice Costs Basic intuitions: • Total costs of deployment

    = sum of all individual service costs • For each service: • Assuming a defined fixed SLA … CostHat
  13. Modelling Microservice Costs Basic intuitions: • Total costs of deployment

    = sum of all individual service costs • For each service: • Assuming a defined fixed SLA … • … costs are a function of load (plus other factors) CostHat
  14. Modelling Microservice Costs Basic intuitions: • Total costs of deployment

    = sum of all individual service costs • For each service: • Assuming a defined fixed SLA … • … costs are a function of load (plus other factors) • “Load” can be internal or external CostHat
  15. Modelling Microservice Costs Basic intuitions: • Total costs of deployment

    = sum of all individual service costs • For each service: • Assuming a defined fixed SLA … • … costs are a function of load (plus other factors) • “Load” can be internal or external • Load changes propagate through the system CostHat
  16. “Costs are a function of load” Data Store Backing Instance

    Backing Instance Backing Instance ... /ep1 /ep2 ... Data Store /ep1 /ep2 ... /ep1 Lambda Function Instance-Backed Microservice Lambda-Backed Microservice (a) (b) API API /ep2 Lambda Function ...
  17. “Costs are a function of load” Data Store Backing Instance

    Backing Instance Backing Instance ... /ep1 /ep2 ... Data Store /ep1 /ep2 ... /ep1 Lambda Function Instance-Backed Microservice Lambda-Backed Microservice (a) (b) API API /ep2 Lambda Function ...
  18. “Costs are a function of load” Data Store Backing Instance

    Backing Instance Backing Instance ... /ep1 /ep2 ... Data Store /ep1 /ep2 ... /ep1 Lambda Function Instance-Backed Microservice Lambda-Backed Microservice (a) (b) API API /ep2 Lambda Function ... EC2, Heroku, Elastic Beanstalk, …
  19. “Costs are a function of load” Data Store Backing Instance

    Backing Instance Backing Instance ... /ep1 /ep2 ... Data Store /ep1 /ep2 ... /ep1 Lambda Function Instance-Backed Microservice Lambda-Backed Microservice (a) (b) API API /ep2 Lambda Function ... EC2, Heroku, Elastic Beanstalk, … AWS Lambda, External Services, …
  20. “Costs are a function of load” Data Store Backing Instance

    Backing Instance Backing Instance ... /ep1 /ep2 ... Data Store /ep1 /ep2 ... /ep1 Lambda Function Instance-Backed Microservice Lambda-Backed Microservice (a) (b) API API /ep2 Lambda Function ... 0 1 2 3 4 5 6 7 8 0 2 4 6 8 10 Requests [1000 per minute] Service Costs [100 $] 10 0 2 3 4 5 6 7 8 9 0 2 4 6 8 Requests [1000 per minute] Service Costs [100 $]
  21. Service Call Graphs frontend /landing 1 customers /login orders /get

    0.2 0.2 recommend /generate 1 orderhistory /get 1 products /get 3.1 10 10 customers /get
  22. Service Call Graphs frontend /landing 1 customers /login orders /get

    0.2 0.2 recommend /generate 1 orderhistory /get 1 products /get 3.1 10 10 customers /get External Call
  23. Service Call Graphs frontend /landing 1 customers /login orders /get

    0.2 0.2 recommend /generate 1 orderhistory /get 1 products /get 3.1 10 10 customers /get
  24. Service Call Graphs frontend /landing 1 customers /login orders /get

    0.2 0.2 recommend /generate 1 orderhistory /get 1 products /get 3.1 10 10 customers /get
  25. Service Call Graphs frontend /landing 1 customers /login orders /get

    0.2 0.2 recommend /generate 1 orderhistory /get 1 products /get 3.1 10 10 customers /get
  26. Case Study Model Building • 7 services, 12 endpoints •

    Combination of instance- and lambda-backed services • Current AWS prices in summer 2016 • Other model parameters: • Assumptions (based on previous benchmarking studies) Customer-Facing Web Interface /get order history /save /generate recommend /get customers /login orders /get /submit banking_ interface /process_ payment /landing frontend /order products /get (focus on what-if analysis)
  27. Four Scenarios Scenario 1 - Improved Conv. Rate • More

    requests to login Scenario 2 - More external load • The recommendation service gets used in other apps as well
  28. Four Scenarios Scenario 1 - Improved Conv. Rate • More

    requests to login Scenario 2 - More external load • The recommendation service gets used in other apps as well
  29. Four Scenarios Scenario 1 - Improved Conv. Rate • More

    requests to login Scenario 2 - More external load • The recommendation service gets used in other apps as well Scenario 3 - Instance Type Migr. • Changing the used backend instance for one service
  30. Four Scenarios Scenario 1 - Improved Conv. Rate • More

    requests to login Scenario 2 - More external load • The recommendation service gets used in other apps as well Scenario 3 - Instance Type Migr. • Changing the used backend instance for one service
  31. Four Scenarios Scenario 1 - Improved Conv. Rate • More

    requests to login Scenario 2 - More external load • The recommendation service gets used in other apps as well Scenario 3 - Instance Type Migr. • Changing the used backend instance for one service Scenario 4 - Code Change • Impact of new dependency to customer/get in the recommendation service
  32. Four Scenarios Scenario 1 - Improved Conv. Rate • More

    requests to login Scenario 2 - More external load • The recommendation service gets used in other apps as well Scenario 3 - Instance Type Migr. • Changing the used backend instance for one service Scenario 4 - Code Change • Impact of new dependency to customer/get in the recommendation service
  33. Related Research Jürgen Cito, Philipp Leitner, Harald C. Gall, Aryan

    Dadashi, Anne Keller, and Andreas Roth. 2015. Runtime metric meets developer: building better cloud applications using feedback. In 2015 ACM International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software (Onward! 2015). ACM, New York, NY, USA, 14-27. DOI=http://dx.doi.org/10.1145/2814228.2814232 R. Singh, P. Shenoy, M. Natu, V. Sadaphal, and H. Vin. Predico: a System for What-If Analysis in Complex Data Center Ppplications. In Proceedings of the 12th International Middleware Conference, pages 120–139, 2011. D. Franceschelli, D. Ardagna, M. Ciavotta, E. Di Nitto. 2013. SPACE4CLOUD: a Tool for System Performance and Cost Evaluation of Cloud Systems. In Proceedings of the 2013 International Workshop on Multi-cloud Applications and Federated Clouds (New York, NY, USA, 2013), MultiCloud ’13, ACM, pp. 27–34.
  34. Related Research Jürgen Cito, Philipp Leitner, Harald C. Gall, Aryan

    Dadashi, Anne Keller, and Andreas Roth. 2015. Runtime metric meets developer: building better cloud applications using feedback. In 2015 ACM International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software (Onward! 2015). ACM, New York, NY, USA, 14-27. DOI=http://dx.doi.org/10.1145/2814228.2814232 Focus on performance rather than costs R. Singh, P. Shenoy, M. Natu, V. Sadaphal, and H. Vin. Predico: a System for What-If Analysis in Complex Data Center Ppplications. In Proceedings of the 12th International Middleware Conference, pages 120–139, 2011. D. Franceschelli, D. Ardagna, M. Ciavotta, E. Di Nitto. 2013. SPACE4CLOUD: a Tool for System Performance and Cost Evaluation of Cloud Systems. In Proceedings of the 2013 International Workshop on Multi-cloud Applications and Federated Clouds (New York, NY, USA, 2013), MultiCloud ’13, ACM, pp. 27–34.
  35. Related Research Jürgen Cito, Philipp Leitner, Harald C. Gall, Aryan

    Dadashi, Anne Keller, and Andreas Roth. 2015. Runtime metric meets developer: building better cloud applications using feedback. In 2015 ACM International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software (Onward! 2015). ACM, New York, NY, USA, 14-27. DOI=http://dx.doi.org/10.1145/2814228.2814232 Focus on performance rather than costs R. Singh, P. Shenoy, M. Natu, V. Sadaphal, and H. Vin. Predico: a System for What-If Analysis in Complex Data Center Ppplications. In Proceedings of the 12th International Middleware Conference, pages 120–139, 2011. D. Franceschelli, D. Ardagna, M. Ciavotta, E. Di Nitto. 2013. SPACE4CLOUD: a Tool for System Performance and Cost Evaluation of Cloud Systems. In Proceedings of the 2013 International Workshop on Multi-cloud Applications and Federated Clouds (New York, NY, USA, 2013), MultiCloud ’13, ACM, pp. 27–34. Palladio
  36. Related Research Jürgen Cito, Philipp Leitner, Harald C. Gall, Aryan

    Dadashi, Anne Keller, and Andreas Roth. 2015. Runtime metric meets developer: building better cloud applications using feedback. In 2015 ACM International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software (Onward! 2015). ACM, New York, NY, USA, 14-27. DOI=http://dx.doi.org/10.1145/2814228.2814232 Focus on performance rather than costs R. Singh, P. Shenoy, M. Natu, V. Sadaphal, and H. Vin. Predico: a System for What-If Analysis in Complex Data Center Ppplications. In Proceedings of the 12th International Middleware Conference, pages 120–139, 2011. What-If Analysis D. Franceschelli, D. Ardagna, M. Ciavotta, E. Di Nitto. 2013. SPACE4CLOUD: a Tool for System Performance and Cost Evaluation of Cloud Systems. In Proceedings of the 2013 International Workshop on Multi-cloud Applications and Federated Clouds (New York, NY, USA, 2013), MultiCloud ’13, ACM, pp. 27–34. Palladio
  37. Conclusions • Costs of changes in Microservice apps are hard

    to predict • CostHat models deployment costs depending on load • Main concepts: • Service call graphs • Cost functions Future work: • Automated monitoring of model parameters • Larger-scale evaluation
  38. Conclusions • Costs of changes in Microservice apps are hard

    to predict • CostHat models deployment costs depending on load • Main concepts: • Service call graphs • Cost functions Future work: • Automated monitoring of model parameters • Larger-scale evaluation