Server Memory - Chernivtsi JS 2019

Transcript

2. @listochkin fb.com/tektonna
3. Not a JavaScript talk
4. No CSS No frameworks No Node
5. A bit deeper than most talks on similar subject
7. How programming languages work
8. Just-in-time compilers Garbage Collection
9. Memory Management
10. Server runs our programs
11. CPU Caches RAM Swap Disk Network
12. Operating System
14. Pages
15. 4k
16. Virtual
17. 2 processes [0 … ∞]
18. Oh, I got a pointer at 4096 What’s at 4095?
19. Page fault
20. Allocate page But not really
21. Once you start writing to it
22. not 4k?
25. malloc
27. malloc me some memory
28. a pointer
29. free(pointer)
30. malloc malloc malloc free malloc free free
31. Forget to free Free the wrong pointer Read from unallocated memory
32. Can we do better?
33. Reference Counting
35. const players = [ … ] players.forEach(p => makeMove(p))
36. a = … doX(a) yield a
37. scope
38. if ( … ) { let a = … … }
40. C++: std::shared_ptr
42. Optimizations
43. Delayed Counting
44. Don’t count local references
45. function (user) { let email = user.email; … … }
46. Delayed deallocation
47. Cycle Collector
50. Compile time ref-counting
51. Several types of pointers
52. Ownership
53. Forbids cycles by forbidding several owning references at the same time
57. Perl 1987 - today Python 1991 - 2001 PHP 1994 - 2009
58. Multiprocess deploys
59. fork
60. Copy on Write
61. Worker Accepts requests As memory use raises: Stop accepting requests Complete in-flight requests Terminate
62. Master Keep track on workers Start new workers Signal them to terminate when memory pressure is high
63. “Pre-fork”
64. Load the framework Load app code Run full GC Start forking process to accept requests
67. “Processes”
68. Tracing GC
69. 1959 60 years ago
70. Start at Root references Follow all references Build a live objects tree Delete all objects not part of the tree
71. Ruby Java JavaScript Lua Go
72. Roots?
73. Constants Global variables Local variables Closures Thread-locals ...
74. Mark & Sweep vs Mark-Compact
76. Can you move objects after mark?
77. Team Sweep: Go Ruby* Lua Embedded JS engines Erlang
78. Pros: Pointers don’t change Native extensions Easier to implement
79. Team Compact: Java JavaScript Ruby* Haskell
80. Pros: Less memory fragmentation over time Cons: Harder Takes longer to do a GC
81. Mark all memory
83. Incremental Marking
84. 3-colored algorithm by Dijkstra™
90. What is a barrier? if statement
91. else branch is very rare CPU branch predictor
92. Parallel marking
95. Lazy Sweep
96. Generational
97. Temporary data
98. Major vs Minor GC
99. Pointers from Old objects to new objects
100. Remembered Set
101. Major GC Scan roots only Bigger object graph
102. Minor GC Scan roots + RS Small object graph
104. Modern GCs are hybrid
107. C-extensions
108. Can’t move objects if their references are passed to an extension
109. Can’t add WB
110. RGen GC
111. 2 types of objects WB-protected WB-unprotected
112. WBu are never OldGen
113. OldGen -> WBu WBu to Remembered Set
114. Mark WBus on every minor GC
115. 1 stw to mark all WBu in RS
118. Adding compaction for WBp
120. V8
121. Minor GC Parallel
122. Major Parallel marking Parallel / Concurrent Compact || Sweep
124. When to trigger GC?
125. Out-of-Bounds GC
126. request? Minor GCs response is sent? Major GC
127. Firefox Run GC in background tabs first instead of current tab
128. Chrome Animation frame
129. Walking the memory
130. Cache locality
131. GPU
132. OS Pages pre-forked processes malloc zones GC Pages Remembered Sets Barriers