Harvest, Yield, and the Promises Our (Storage) Services Make

An overview of Harvest and Yield from Fox and Brewer's "Harvest, Yield, and Scalable Tolerant Systems" and a discussion about how infrastructure developers communicate trade-offs to their users.

Transcript

1. HARVEST, YIELD, AND THE PROMISES OUR (STORAGE) SERVICES MAKE Jordan West - Distributed Reading Group 2/25/16
3. AGENDA Context Harvest & Yield: Definition Harvest & Yield: Back to CAP Harvest & Yield: Criticism Discussion
4. THE PAPER IN CONTEXT
5. 1999 Same year “Harvest, Yield, and Scalable Tolerant Systems” was published
6. TALK CONTEXT
8. - Fox & Brewer “In practice, many applications are best described in terms of reduced consistency or availability”
9. CAP SAYS VERY LITTLE ABOUT A SYSTEM
10. THAT DOESN’T MEAN CAP SHOULD BE DISREGARDED
11. GOAL: ENCOURAGE DISCUSSION OF BETTER WAYS TO COMMUNICATE
12. HARVEST & YIELD
13. - Fox and Brewer “Yield…is the probability of completing a request”
14. - Fox and Brewer “Harvest…measures the fraction of data reflected in the response…”
15. - Fox & Brewer “The model can be applied in the case of ‘single- location’ updates..that affect reachable nodes occur correctly but have limited visibility (a form of reduced harvest), while those that require unreachable nodes fail (reducing yield)”
16. – Fox and Brewer “In the presence of faults there is typically a tradeoff between providing no answer (reducing yield) and providing an imperfect answer (maintaining yield, but reducing harvest).”
17. Y I E L D H
 A R V E S T 0 100% 100%
18. Y I E L D H
 A R V E S T Strong Consistency 0 100% 100%
19. Y I E L D H
 A R V E S T Strong Consistency 0 100% 100%
20. Y I E L D H
 A R V E S T Strong Consistency Eventual Consistency 0 100% 100%
21. Y I E L D H
 A R V E S T Strong Consistency Eventual Consistency 0 100% 100%
22. Y I E L D H
 A R V E S T Eventual Consistency 0 100% 100%
23. Y I E L D H
 A R V E S T 0 100% 100%
24. Y I E L D H
 A R V E S T Cassandra 0 100% 100%
25. Y I E L D H
 A R V E S T Cassandra 0 100% 100%
26. Y I E L D H
 A R V E S T Cassandra 0 100% 100%
27. Y I E L D H
 A R V E S T Cassandra Riak 0 100% 100%
28. Y I E L D H
 A R V E S T Cassandra Riak 0 100% 100%
29. Y I E L D H
 A R V E S T Cassandra Riak 0 100% 100%
30. Y I E L D H
 A R V E S T Cassandra Riak 0 100% 100% Writes & Reads via Consensus
31. Y I E L D H
 A R V E S T Cassandra Riak 0 100% 100% Writes & Reads via Consensus
32. Y I E L D H
 A R V E S T Cassandra Riak 0 100% 100% Writes & Reads via Consensus Writes via Consensus, Read Any Replica
33. Y I E L D H
 A R V E S T Cassandra Riak 0 100% 100% Writes & Reads via Consensus Writes via Consensus, Read Any Replica
34. SO CAP….
35. Y I E L D H
 A R V E S T 0 100% 100%
36. Y I E L D H
 A R V E S T All CAP Says is “Can’t have both 100% Harvest and 100% Yield” 0 100% 100%
37. Y I E L D H
 A R V E S T All CAP Says is “Can’t have both 100% Harvest and 100% Yield” 0 100% 100% Can Have: CP
38. Y I E L D H
 A R V E S T All CAP Says is “Can’t have both 100% Harvest and 100% Yield” 0 100% 100% Can Have: CP Or Have: AP
39. NOTHING IS PERFECT
41. THAT PESKY CA
42. WHAT ABOUT LATENCY?
43. PACELC
44. – Abadi “if there is a partition (P), how does the system trade off availability and consistency (A and C); else (E), when the system is running normally in the absence of partitions, how does the system trade off latency (L) and consistency (C).”
45. - Fox, Brewer and West (Kingsbury) “Yield…is the probability of completing a request within an acceptable latency bound”
46. “SINGLE-LOCATION” UPDATES
47. HARVEST IS DIFFICULT TO MEASURE
48. DISCUSSION