1 / 54

Relativistic Red-Black Trees

Relativistic Red-Black Trees. Philip Howard 10/31/2011 phil.w.howard@gmail.com. Red-Black Trees. External Property : Every external node is black Internal Property : The children of red nodes are black Depth Property : All external nodes have the same black depth. 12. 11. 15. 5. 11.5.

hamlet
Télécharger la présentation

Relativistic Red-Black Trees

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Relativistic Red-Black Trees Philip Howard 10/31/2011 phil.w.howard@gmail.com

  2. Red-Black Trees • External Property: Every external node is black • Internal Property: The children of red nodes are black • Depth Property: All external nodes have the same black depth

  3. 12 11 15 5 11.5 13 17 11.2 11.7 3 9 14 4 7 10 6

  4. Reasons for RBTrees • O( log(n) ) performance • O(1) restructures after insert/delete • Possibly O( log(n) ) recolors (but recolors are cheap)

  5. Insertion • Search until we get to external node • Insert at this location adding two empty, black, external nodes underneath • If root, color black, else color red • Preserves external, and depth properties; may violate internal property

  6. 12 15 5 13 17 3 10 14 4 7 11 6 8 9 Insert 9

  7. Delete • Find node • If node doesn’t have an external child, swap with next node in in-order traversal order (left most node of the right branch) • Delete the node from the bottom

  8. Swap example: Delete 5 12 11 15 5 11.5 13 17 11.2 11.7 3 8 14 4 6 10 7

  9. Diag Restructure C B B 4 A C A 3 2 3 4 1 1 2

  10. Zig Restructure C B A 4 A C B 1 2 3 4 1 2 3

  11. Performance • Find, Insert, Remove O( log(n) ) • At most 1 restructure on insert • At most 2 restructures on delete • Possibly log(n) recolors

  12. synchronize_rcu rcu_assign_pointer rcu_dereference call_rcu call_rcu(free) wait-for-readers rp-publish rp-read defer-for-readers rp-free Terminology

  13. Terminology Sn the Start of operation n. Fn the Finish of operation n. En the Effect of operation n. For example, if operation n is an insert, the effect of that operation is that the tree has a new node. ab defines a happens-before relation such that a happens before b.

  14. Terminology if SaSb Fa or Sb Sa Fb then a and b are concurrent and either EaEb or Eb Ea is possible

  15. Relativistic Requirements • Readers don’t block, and are not inhibited by writers • Lookups always find a node that exists in the tree • Traversals always return nodes in the correct order without skipping any nodes that exist in the tree

  16. nodes that “exist in the tree” i is insert of N d is delete of N r is read (lookup or complete traversal) • If FiSr and Fr Sd then N exists in the tree • If Fr Si or Fd Sr then N does not exist in the tree

  17. nodes that “exist in the tree” i is insert of N d is delete of N r is read (lookup or complete traversal) • If i is concurrent with r then either EiEr or Er Ei • If d is concurrent with r then either EdEr or Er Ed

  18. nodes that “exist in the tree” • Any update that strictly precedes a read is observable by that read • Any update that strictly follows a read is not observable by that read • Any update that is concurrent with a read may or may not be observable by that read

  19. Relativistic Implementation

  20. Assumptions • Readers ignore the color of nodes • Readers don’t need to access the parent pointers (note: some traversal algorithms require access to parent pointers) • If ADT is a map (not a multimap), then temporarily having a duplicate node in the tree won’t affect lookups • Mutual exclusion used between writers

  21. Operations Insertion No nodes are moved or freed so this operation is safe. Recolor Recoloring does not affect the read consistency of the tree. Delete Need to defer reclamation of memory until no thread has a reference to the deleted node. Swap Need to guarantee that the node which moves up isn’t missed by readers between the new and old position. Restructures Requires moving nodes. Need to ensure that no readers get lost in the transition.

  22. Diag Restructure H H F D D G B F B E A C E G A C

  23. Diag Restructure H H H F F F D G D G D G B B B E E E A C A C A C

  24. Diag Restructure H H F D D G B F B E A C E G A C Reader at F looking for D?

  25. Diag Restructure H H H F F D G D D B E B F’ B F’ A C A C E G A C E G • init F’ • link F’ • unlink F

  26. RBTree Delete with Swap F F F B C’ B C’ A E E A E A C C D D null null D

  27. Swap algorithm find node to delete find swap candidate create swap-prime link swap-prime into tree wait a grace period remove swap from tree

  28. Diag Restructure H H F D D G B F B E A C E G A C

  29. RBTree Delete with Swap F F F B C’ B C’ A E E A E A C C D D null null D

  30. Zig Restructure C B A 4 A C B 1 2 3 4 1 2 3

  31. Performance nolock No synchronization – NOT A VALID IMPLEMENTATION lock pthread mutex lock rwlr Reader-Writer lock that favors readers rwlw Reader-Writer lock that favors writers rp Relativistic Programming implentation ccavl Concurrent AVL implementation

  32. Read Performance

  33. Contended Write Performance

  34. Linearizability?

  35. Linearizability • Definition Equivalent to a some valid serial execution • Proof Each operation appears to take effect instantaneously

  36. RP RBTree Properties • Immutable nodes – key & value don’t change • Sort – Nodes are always in valid sort order • Structural Integrity – lookups will always succeed

  37. Lookup Three options at each node • correct node was found • proceed down left path • proceed down right path

  38. Lookup Last rp-read is the linearization point • Immutable nodes means each time we examine a node, we’ll make the same decision • Sort means we’ll make the right decision at each node • Structural integrity means concurrent updates won’t break our lookup Must show that updates maintain these properties

  39. Insert • rp-publish is the linearization point • Must show that restructure does not invalidate sort or structural integrity properties

  40. Delete without Swap • rp-publish is the linearization point (assigning NULL or the single child to the parent of deleted node) • Must show that restructure does not invalidate sort or structural integrity properties

  41. F C’ F B F F C’ B E A C’ A E C A E C null A E C D D D null D Swap • rp-publish(C’) is linearization point • Consider linearization point of lookups for B • Preservation of properties?

  42. Diag Restructure D D D C C B 4 B B A 3 A C’ A C’ 1 2 1 2 3 4 1 2 3 4 • Linearization point? • sort? • structural integrity?

  43. Linearizability? Write 1 Write 2 Reader

  44. Traversals 12 11 15 5 11.5 13 17 11.2 11.7 3 9 14 4 7 10 6

  45. Traversal Options • Use a stack to represent current location in traversal • Use parent pointers to track location in traversal • Other options?

  46. Lookup AssumptionsValid for Traversals? • Readers ignore the color of nodes • Readers don’t need to access the parent pointers (note: some traversal algorithms require access to parent pointers) • If keys are unique, temporarily having a duplicate node in the tree won’t affect reads • Mutual exclusion used between writers

  47. Traversal Options • Reader-Writer lock O(N) traversal – with delayed writes • Relativistic O( N log(N) ) traversal – with undelayed writes • Relativistic O(N) traversal – with more complicated writes

  48. Contended Update Performance

  49. Read Performance 64K node tree

  50. Scalability (4 readers)

More Related