1. Ana-Maria Oprescu, Thilo Kielmann (Vrije University) Presented By Gal Cohen Cloud Computing Seminar CS Technion, Spring 2012

2.  High throughput computing jobs  No interactive deadline  Tasks are independent of each other  All tasks are ready for execution  Unknown runtimes  Execution Model: ◦ Allocate resources (e.g. machines) ◦ Run each task (once) from the bag on some machine 2

3.  Unknown runtime distribution  However, some distribution exists  The total number of jobs is also known  Tasks can be aborted 3

4.  There are many Cloud providers. (EC2, Azure, Rackspace, 3Tera)  Many types of machines even in the same provider, for a different price. ◦ CPU count and speed ◦ Memory size  Upper limit on the number of machines assignable from a provider (self imposed)  A machine is charged per ATU (Hour) 4

5.  The Goal ◦ Run all the tasks from a given bag on cloud computers, meeting a limited budget ◦ Minimize the makespan of the whole bag (without exceeding the budget constraint)  Assumption ◦ Running each task on a machine separately (FIFO) 5

6.  The scheduler (BaTS) runs outside of the cloud (for free)  The scheduler gets the Bag Of Tasks  It allocates machines from each cloud  Dispatch jobs to the allocated machines  Receives feedback on tasks completion 6

7. 7

8. 8

9. 9 Error Level Typical Values: 0.10,0.15,0.20,0.25

10. 10 Required sample size (n) Bag Of Tasks Size (N)

11. 11

12. 12

13. 13

14. 14 ATU cost for machine of type i

15. 15

16. 16

17. 17

18. 18

19. 19

20. 20

21.  Thus, BaTS continuously tries to avoid budget violations  Theoretically, It’s easy. As the execution continues, the bag is smaller and the budget is smaller.  The trouble is estimating the size of the bag at a given moment. (some machines will finish their current job before ATU ends) 21

22. 22

23. 23

24. 24

25. 25

26. 26

27. 27

28. 28

29.  “Machine speed” in each “cloud” was simulated according to 5 scenarios: 29 Profitability C2 w.r.t C1 Cloud 2 CostSpeed 1/441 3/443 111 4/334 414

30. In each scenario, comparing RR to BaTS  RR always uses 32+32 machines  BaTS initial configuration is 30+30 machines and ◦ Budget B = the cost of running RR for that scenario ◦ Budget B = the cost of running only on the most “profitable” machine type. (computed offline) 30

31. 31

32. 32

33. 33

34.  BaTS helps choosing the cloud resources suitable for an application  BaTS helps scheduling within budget while still performing reasonably well 34

35.  Limitations ◦ The provided tests “cheat” because the number of machines is very small ◦ The “Tail phase” is not handled well (The “faster” machines will be released before the “slow” ones) ◦ Guessing a proper budget ◦ Actual Bags on actual clouds ◦ What about data transfer costs? ◦ Storage constraints? ◦ Other metric – maximize the profitability (or minimize the budget) while not exceeding a given makespan 35