The Potential Dangers of Causal Consistency and an Explicit Solution

http://www.bailis.org/papers/explicit-socc2012.pdf
SOCC 2012

Transcript

1. the potential dangers of causal consistency and an explicit solution Peter Bailis, Alan Fekete, Ali Ghodsi, Joseph M. Hellerstein, Ion Stoica SOCC 2012
2. A Story
3. A Story
4. “Sally’s in a coma!” A Story
5. “Sally’s in a coma!” “Sally’s in a coma!” A Story “Sally’s okay!”
6. “Sally’s in a coma!” “Sally’s in a coma!” A Story “Sally’s okay!”
7. “Sally’s in a coma!” “Sally’s in a coma!” A Story “Great news!” “Sally’s okay!”
8. “Sally’s in a coma!” “Sally’s in a coma!” A Story “Great news!” “Sally’s okay!”
9. “Sally’s in a coma!” “Sally’s in a coma!” “Great news!” “Sally’s okay!” The Wrong Story
10. “Sally’s in a coma!” “Great news!” The Wrong Story
11. “Sally’s in a coma!” “Great news!” The Wrong Story
12. “Sally’s in a coma!” “Great news!” The Wrong Story
13. “Sally’s in a coma!” “Great news!” The Wrong Story ???
14. The Right Order “Sally’s in a coma!” “Great news!” “Sally’s okay!”
15. The Right Order “Sally’s in a coma!” “Great news!” “Sally’s okay!”
16. The Right Order “Sally’s in a coma!” “Great news!” “Sally’s okay!”
17. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
18. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
19. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
20. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
21. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
22. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
24. 1978 “Time, Clocks, and the Ordering of Events in a Distributed System” by Leslie Lamport
25. 1978 “Time, Clocks, and the Ordering of Events in a Distributed System” by Leslie Lamport 1994 “Causal Memory” by Ahamad et al.
26. 1978 “Time, Clocks, and the Ordering of Events in a Distributed System” by Leslie Lamport 1994 “Causal Memory” by Ahamad et al. 2011-12
27. 2011-12
28. 2011-12
29. 2011-12 IEEE CAP There are many hard, trade-offs,
30. 2011-12 IEEE CAP SOSP: COPS There are many hard, trade-offs, but causal consistency can work well
31. 2011-12 IEEE CAP SOSP: COPS Texas: CAC There are many hard, trade-offs, but causal consistency can work well and it’s the best you can do*!
33. “Sally’s okay!”
34. “Sally’s okay!”
35. “Sally’s okay!”
36. “Sally’s okay!”
37. “Sally’s okay!”
38. “Sally’s okay!”
39. “Great news!” “Sally’s okay!”
40. “Great news!” “Sally’s okay!”
41. “Great news!” “Sally’s okay!”
42. “Great news!” “Sally’s okay!”
46. Deliver writes in the “right” order
47. Deliver writes in the “right” order “Great news!” “Sally’s okay!” “Sally’s in a coma!”
48. Causal Consistency reads obey partial order called “happens-before”
58. Write value
59. Write value Happens-before metadata
60. Write value Happens-before metadata
64. Have we seen all dependencies?
65. Have we seen all dependencies?
66. Have we seen all dependencies? Y
67. Have we seen all dependencies?
68. Have we seen all dependencies? Y
69. Have we seen all dependencies? Yes: apply write
70. Have we seen all dependencies? Yes: apply write
71. Have we seen all dependencies? No: wait Yes: apply write
72. Sounds great, right?
73. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
74. Causal Consistency reads obey partial order called “happens-before”
75. Causal Consistency traditional happens-before: potential causality reads obey partial order called “happens-before”
76. If you wish to make an apple pie from scratch, you must first invent the universe. -Carl Sagan
78. “Great news!”
79. “Great news!” “Sally’s okay!”
80. “Great news!” “Sally’s okay!” “Great picture!” “Rad party!” “Lol!” “Can’t wait for SOCC!” “Want to go skiing?”
81. “Great news!” “Sally’s okay!” “Great picture!” “Rad party!” “Lol!” “Can’t wait for SOCC!” “Want to go skiing?” “I hope my paper gets in.” “You who?” “I love Tahoe!” “Coming tonight?
82. “Great news!” “Sally’s okay!” “Great picture!” “Rad party!” “Lol!” “Can’t wait for SOCC!” “Want to go skiing?” “Snow rocks!” “I hear the PC is great!” “You.” “Hello, “I hope my paper gets in.” “You who?” “I love Tahoe!” “Coming tonight?
83. “Great news!” “Sally’s okay!” “Great picture!” “Rad party!” “Lol!” “Can’t wait for SOCC!” “Want to go skiing?” “Snow rocks!” “I hear the PC is great!” “You.” “Hello, “Who’s there?” “Are you submitting?” “Have you met Larry?” “Great food here” “I hope my paper gets in.” “You who?” “I love Tahoe!” “Coming tonight?
84. “Great news!” “Sally’s okay!” “Great picture!” “Rad party!” “Lol!” “Can’t wait for SOCC!” “Want to go skiing?” “Snow rocks!” “I hear the PC is great!” “You.” “Hello, “Who’s there?” “Are you submitting?” “Have you met Larry?” “Great food here” “Knock, knock.” “I hope my paper gets in.” “You who?” “I love Tahoe!” “Coming tonight?
85. “Great news!” “Sally’s okay!” “Great picture!” “Rad party!” “Lol!” “Can’t wait for SOCC!” “Want to go skiing?” “Snow rocks!” “I hear the PC is great!” “You.” “Hello, “Who’s there?” “Are you submitting?” “Have you met Larry?” “Great food here” “Knock, knock.” “I hope my paper gets in.” “You who?” “I love Tahoe!” “Coming tonight?
86. Twitter.com 20 Tweets at login 20 more auto-scroll 600+ per min
87. Twitter.com quickly approaches asymptotic limit
88. r e c a p
89. potential danger Potential causality graphs are huge, limiting local apply rates and throughput r e c a p
90. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
96. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER!
97. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! Unstable queue!
98. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! Unstable queue! Sustained throughput limited to slowest DC
99. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! Unstable queue! Sustained throughput limited to slowest DC
100. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! Sustained throughput limited to slowest DC
101. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! Adding DCs doesn’t help slowest site Sustained throughput limited to slowest DC
102. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! During partitions/failures, sustainable throughput is zero writes/s zero Sustained throughput limited to slowest DC
103. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! During partitions/failures, sustainable throughput is zero writes/s zero
104. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! During partitions/failures, sustainable throughput is zero writes/s zero
105. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! During partitions/failures, sustainable throughput is zero writes/s zero Metadata garbage collection stalls stalls
106. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! During partitions/failures, sustainable throughput is zero writes/s zero Metadata garbage collection stalls stalls
107. DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! DANGER! During partitions/failures, sustainable throughput is zero writes/s zero Metadata garbage collection stalls stalls
108. r e c a p
109. r e c a p potential danger Write throughput limited to slowest DC Violation 㱺 arbitrarily high visibility latency Adding DCs does not increase throughput
110. Potential dangers spell trouble for causality bad news
111. Potential dangers spell trouble for causality to it? Are live without we destined bad news
112. Potential dangers spell trouble for causality to it? Are live without we destined bad news
113. SOME DEPENDENCIES METADATA AND MODERN SERVICES (NOT) SCALING UP AND OUT BETTER LIVING VIA APP CONTEXT
114. Observation Why track dependencies that don’t matter?
115. Observation Why track dependencies that don’t matter?
116. Observation Why track dependencies that don’t matter? Need to know application semantics
117. Explicit Causality app-defined “happens-before” transitivity enforced subset of potential causality
118. Explicit Causality app-defined “happens-before” transitivity enforced subset of potential causality not a new idea (e.g., Cheriton and Skeen SOSP 1993, Ladin et al. PODC 1990) but...
119. Explicit Matters
120. Explicit Matters Twitter 28% of Tweets in conversations 69% of convos are depth two average depth is 10.7 [Ye and Wu SocInfo 2010, Ritter et al. HLT 2010]
121. Explicit Matters Twitter 28% of Tweets in conversations 69% of convos are depth two average depth is 10.7 [Ye and Wu SocInfo 2010, Ritter et al. HLT 2010] reply-to degree and depth are limited
122. Explicit Matters Twitter 28% of Tweets in conversations 69% of convos are depth two average depth is 10.7 [Ye and Wu SocInfo 2010, Ritter et al. HLT 2010] reply-to degree and depth are limited 109 smaller graph for a year of Tweets
123. Explicit API put(key, value)
124. put_after(key, value, deps) Explicit API
125. put_after(key, value, deps) Explicit API
126. put_after(key, value, deps) Explicit API (possibly empty) set of references to other writes
127. put_after(key, value, deps) Explicit API track what matters (possibly empty) set of references to other writes
128. put_after(key, value, deps) Explicit API track what matters frequently in data model already (possibly empty) set of references to other writes
129. put_after(key, value, deps) Explicit API track what matters frequently in data model already can simulate fencing (possibly empty) set of references to other writes
130. put_after(key, value, deps) Explicit API track what matters frequently in data model already can simulate fencing (possibly empty) set of references to other writes won’t track non-explicit references
131. NATURAL EXTENSION NOT ROCKET SCIENCE
132. NATURAL EXTENSION NOT ROCKET SCIENCE http://twitter.com/rbranson/status/256795094387142657
134. potential causality
136. A WORLD OF PHYSICAL OPERATIONS MESSAGES REGISTERS READS WRITES MODEST SCALE
139. A WORLD OF HUGE SCALE MASSIVE INTERACTIVITY PLANET-WIDE NETWORKS REAL-WORLD INTERFACES UBICOMP BIG DATA
141. potential dangers
142. potential dangers huge causality graphs
143. potential dangers huge causality graphs throughput scalability limited
144. potential dangers huge causality graphs explicit causality semantic context to the rescue consider modern apps helps with #1, indirectly with #2 throughput scalability limited