1. Capsule Placement in the Service Platform Bhuvan Urgaonkar Timothy Roscoe Systems Group, Sprint ATL

2. Service Platform: an overview Processors High speed interconnect Internet Management/Control Unit

3. Service Platform: Goals Sell the platform’s resources Manage the resources efficiently Provide performance guarantees to customers Start or stop services within minutes

4. Services and Capsules Services: –web/game/streaming servers –service provider pays the platform Capsules –Def: Component of a service that should run on a single node –e.g.: consider a replicated web server

5. Nucleus Node specific control/management software: –Capsule creation, destruction –Health information (process liveness) –Resource parameters (memory, CPU, network bandwidth etc.)

6. Control Plane Capsule Placement Flow Placement Node, network, service monitoring Deployed Service Database Billing

7. Outline of this talk Service Platform: an overview Quality of Service Capsule Placement Design of the Placement unit Conclusions and future work

8. QoS Representations Application level –e.g., 50 transactions per sec Contract level –e.g., “something like a 300 MHz Pentium II” Platform level –e.g., ? Node level –e.g., weights, priorities etc.

9. Translation between QoS levels Application level => Contract level –Application specific, customer’s problem Contract level => Platform level –More a business problem Platform level => Node level –OS dependent

10. Capsule Placement: Desirables Maximize revenue! Aware of the “importance” of services. Overbooking. Exploit known workload characteristics. Adapt to changes in workload? Fast.

11. Stages in hosting a service Requirement specification Placement Deployment Activation

12. Requirement Specification Contract level representation –Many possibilities: 300 MHz PII, best effort or a CPU instruction token bucket. Platform level representation –Must be uniform across the platform. –(rate, burst, ovb tolerance, arch, OS)

13. Translation to Node level Reservation based scheduler –map (rate, burst) to (period, slice) –bigger burst => bigger period Proportional share scheduler –burst ? –weight in proportion to rate Priority based scheduler –no easy mapping

14. Placement Find the set of feasible nodes –Compatible architecture and OS –No overbooking tolerances violated Pick one node from this set –Best Fit –Worst Fit –Random Select –Close Overbooking

15. Placement: Example capsules nodes abc N1N4N3N2 One possible placement: (a, N1), (b, N2), (c, N3) 103020 30102010

16. Deployment and Activation Deployment: The process of preparing a capsule for execution on a node. –Why ? e.g., need to download some files before starting –the control plane sends all information to deploy the capsule Activation: Starting a deployed service

17. Capsule State Diagram deployedundeployed deploying undeploying activating deactivating active

18. Example Message Exchange Control PlaneNucleus deployed svc cap state svc cap deployed deployed svc cap Instruct nucleus to deploy a capsule, start timer No response! Send again Starts deploying the capsule Still deploying Done deploying, send status message Deployed before timeout, instruct nucleus to activate activated svc cap Starts activating the capsule...

19. Placement Unit Architecture Listen for new requests Event QueueMessage Queue Dispatch EventsListen to nuclei Events due to new requests Events due to msgs from nucleiMessages from nuclei

20. Database Consistency Transactions and exceptions –e.g: try: transaction_begin () deploy_service (svc): transaction_commit () except: transaction_abort ()

21. Performance Time to compute placement: 1-2 sec => time to deploy usually much larger Comparison of heuristics –experiments with following workloads 1-3 capsules, CPU requirement 0-10%, wide range of overbooking tolerances –Random Select admitted most # services, Best Fit admitted least –But … more investigation needed

22. Summary QoS representation for CPU requirements of services. Implementation of placement unit. Some simple experiments to deploy and activate services.

23. Unfinished... Experiments: –heuristics better suited to specific workloads. –Scalability and efficiency of the system. Integration of placement unit with rest of the Control Plane Handling various failures Extend to multiple resources - much harder than a single resource!