1 / 33

Multi-resource Allocation with Unknown Participants

Multi-resource Allocation with Unknown Participants. Ajoy K. Datta , Stéphane Devismes, François Kawala , Lawrence L. Larmore , and Maria Potop-Butucaru. K-resource Allocation. C 1. N >> M. C 2. R 1. C 3. R 2. C 4. …. C 5. R M. C 6. …. Resources. C N. Clients.

gilda
Télécharger la présentation

Multi-resource Allocation with Unknown Participants

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. Multi-resource Allocation with Unknown Participants Ajoy K. Datta, Stéphane Devismes, François Kawala, Lawrence L. Larmore, and Maria Potop-Butucaru

  2. K-resource Allocation C1 • N >> M C2 R1 C3 R2 C4 … C5 RM C6 … Resources CN PDAA'2011, Osaka Clients

  3. K-resource Allocation C1 • N >> M • Clients don’t know each other C2 R1 C3 R2 C4 … C5 RM C6 … Resources CN PDAA'2011, Osaka Clients

  4. K-resource Allocation C1 R1,R2 • N >> M • Clients don’t know each other • Request: up to K resources • Clients only know the IDs of the resource they request • Request given by an oracle C2 R1 C3 R1,R3 R2 C4 R4 … C5 RM R2, R6,R7 C6 … Resources CN PDAA'2011, Osaka Clients

  5. K-resource Allocation C1 R1,R2 C2 R1 C3 R1,R3 R2 C4 R4 … C5 RM R2, R6,R7 C6 • SAFETY: Each resource is used by at most one client at a time … Resources CN PDAA'2011, Osaka Clients

  6. K-resource Allocation C1 R1,R2 C2 R1 C3 R1,R3 • LIVENESS: • Every request is eventually satisfied R2 C4 R4 … C5 RM R2, R6,R7 C6 … Resources CN PDAA'2011, Osaka Clients

  7. Related Work • K-out-of-L Exclusion • Drinking philosophers • Application: Peer-to-Peer systems PDAA'2011, Osaka

  8. Model • Asynchronous message passing • Reliable link • Any client can only communicate with the resources it requests PDAA'2011, Osaka

  9. 2-Resource Allocation • Overview • Queues • At each resource • To store client’s requests • The request at head of the queue is satisfied • Issue • Deadlock C2 C1 C3 R5 PDAA'2011, Osaka

  10. 2-Resource Allocation C1 C2 R1 C3 R2 R3 PDAA'2011, Osaka

  11. Deadlock C1 C2 C1 C2 R1 C3 C3 C2 C1 C3 R2 R3 PDAA'2011, Osaka

  12. Our solution • First request: strong • (Second request: weak) • Two queues per resource: strong, weak • Resource allocated to the head of its strong queue • Weak requests move from weak to strong queue • Under some conditions PDAA'2011, Osaka

  13. Our solution C1 R1 C1,R1,Weak R2 PDAA'2011, Osaka

  14. Our solution C1 R1 C1,R2,Strong C1 R2 PDAA'2011, Osaka

  15. Our solution C1 C1 R1 C1 R2 PDAA'2011, Osaka

  16. Our solution C1 StrongReady C1 R1 C1 R2 PDAA'2011, Osaka

  17. Our solution C1 C1 R1 C1 R2 PDAA'2011, Osaka

  18. Our solution C1 C1 R1 C1 ResAllowed R2 PDAA'2011, Osaka

  19. Our solution C1 C1 R1 CS C1 R2 PDAA'2011, Osaka

  20. Our solution C1 C1 R1 Done C1 R2 PDAA'2011, Osaka

  21. Our solution C1 C1 R1 Done R2 PDAA'2011, Osaka

  22. Our solution EndAck C1 R1 R2 PDAA'2011, Osaka

  23. Deadlock C1 C3 C2 C1 C3 C2 R1 R2 R3 PDAA'2011, Osaka

  24. Deadlock C2 C3 C1 C3 C1 C2 R1 R2 R3 PDAA'2011, Osaka

  25. Dependancy Cycle C2 C3 C1 C3 C1 C2 R1 R2 R3 PDAA'2011, Osaka

  26. Conflict Resolution • Dependency cycle detection • A message follows the dependencies • Dependency cycle breaking • A dependency is broken • A client’s request is penalized PDAA'2011, Osaka

  27. Fairness • Penalization must be fair • Identifiers cannot be used • We use a token • circulating on the resource ring PDAA'2011, Osaka

  28. Token (1/2) • Token reception • The resource marks all its strong requests • Token releasing • When all marked request have been satisfied • Forwarded to the next resource in the ring • Each resource gets the token infinitely often PDAA'2011, Osaka

  29. Token (2/2) • Token holder never penalized • Penalized dependency: • the one out-coming from the smallest non token holder • The token holder “flushes” its “old” requests before releasing the token PDAA'2011, Osaka

  30. Dynamicity • Join • New identity • Leave • With announcement: easy • Crash • Need a participant detector PDAA'2011, Osaka

  31. Participant Detector • Required : Perfect [Fetzer,2003] • Strong completeness: Every client that leaves tge system is eventually removed from the participant lists • Strong accuracy: No client can be removed from a list before it leaves the system PDAA'2011, Osaka

  32. K > 2 • Generalized the previous solution: hard • Pessimistic approach: prevent deadlock creation • Resource allocated sequentially • Not efficient (not enough concurrency) • Hybrid solution ? PDAA'2011, Osaka

  33. Thank you PDAA'2011, Osaka

More Related