First year PhD review

Transcript

1. Optimising Concurrency Control for Distributed applications Sreeja S Nair First year review 24th June 2019 Advised by Marc Shapiro
2. Distributed Applications • Modern day applications serve users in different geographical regions • Applications are replicated closer to users due to demands of low response time • CAP theorem depicts a tension between Consistency and Availability 2
3. Replicated Online Auction 3 Alice Bob Charles
4. Replicated Online Auction 3 Alice Bob Charles
5. Replicated Online Auction 3 Alice Bob Charles
6. Replicated Online Auction 3 Alice Bob Charles
7. Replicated Online Auction 3 Alice Bob Charles
8. Replicated Online Auction 3 Alice Bob Charles
9. Replicated Online Auction 3 Alice Bob Charles
10. Replicated Online Auction 3 Alice Bob Charles
11. Replicated Online Auction 3 Alice Bob Charles
12. • CISE • Alexey Gotsman, Hongseok Yang, Carla Ferreira, Mahsa Najafzadeh, and Marc Shapiro. 'Cause I'm strong enough: Reasoning about consistency choices in distributed systems. POPL ‘16 • CEC • Gonçalo Marcelino, Valter Balegas, and Carla Ferreira. 2017. Bringing Hybrid Consistency Closer to Programmers. PaPoC ’17 • Sreeja Nair, Marc Shapiro. Improving the "Correct Eventual Consistency" Tool. [Research Report] RR-9191, Sorbonne Université. 2018. • Hamsaz • Farzin Houshmand and Mohsen Lesani. 2019. Hamsaz: replication coordination analysis and synthesis. POPL ‘19 • Soteria • Sreeja Nair, Gustavo Petri, Marc Shapiro. Invariant Safety for Distributed Applications. PaPoC ’19 • Sreeja Nair, Gustavo Petri, Marc Shapiro. Proving the safety of highly-available distributed objects, submitted to OOPSLA ‘19 Proving safety of distributed applications 4
13. • CISE • Alexey Gotsman, Hongseok Yang, Carla Ferreira, Mahsa Najafzadeh, and Marc Shapiro. 'Cause I'm strong enough: Reasoning about consistency choices in distributed systems. POPL ‘16 • CEC • Gonçalo Marcelino, Valter Balegas, and Carla Ferreira. 2017. Bringing Hybrid Consistency Closer to Programmers. PaPoC ’17 • Sreeja Nair, Marc Shapiro. Improving the "Correct Eventual Consistency" Tool. [Research Report] RR-9191, Sorbonne Université. 2018. • Hamsaz • Farzin Houshmand and Mohsen Lesani. 2019. Hamsaz: replication coordination analysis and synthesis. POPL ‘19 • Soteria • Sreeja Nair, Gustavo Petri, Marc Shapiro. Invariant Safety for Distributed Applications. PaPoC ’19 • Sreeja Nair, Gustavo Petri, Marc Shapiro. Proving the safety of highly-available distributed objects, submitted to OOPSLA ‘19 State-propagating applications Proving safety of distributed applications 4 Operations-propagating applications Operations-propagating applications Operations-propagating applications
14. • CISE • Alexey Gotsman, Hongseok Yang, Carla Ferreira, Mahsa Najafzadeh, and Marc Shapiro. 'Cause I'm strong enough: Reasoning about consistency choices in distributed systems. POPL ‘16 • CEC • Gonçalo Marcelino, Valter Balegas, and Carla Ferreira. 2017. Bringing Hybrid Consistency Closer to Programmers. PaPoC ’17 • Sreeja Nair, Marc Shapiro. Improving the "Correct Eventual Consistency" Tool. [Research Report] RR-9191, Sorbonne Université. 2018. • Hamsaz • Farzin Houshmand and Mohsen Lesani. 2019. Hamsaz: replication coordination analysis and synthesis. POPL ‘19 • Soteria • Sreeja Nair, Gustavo Petri, Marc Shapiro. Invariant Safety for Distributed Applications. PaPoC ’19 • Sreeja Nair, Gustavo Petri, Marc Shapiro. Proving the safety of highly-available distributed objects, submitted to OOPSLA ‘19 State-propagating applications Proving safety of distributed applications 4 Operations-propagating applications Operations-propagating applications Operations-propagating applications Work during year 1 Work during year 1
15. Auction application • State • Sellers • Buyers • Products • Auctions • Bids 5 • Operations • Register and unregister sellers and buyers • Adding and removing products • Creating, removing, starting and closing an auction • Adding and removing lots in an auction • Placing and removing bids • Invariants • A seller has a limit on the number of items they can sell • The stock for each seller has to be non-negative • The highest bid wins
16. Analysing online auction • Place bid - Close auction needs to synchronise to ensure the highest bid wins • Concurrency control options: • Mutual exclusion - total order for both operations • Async/wait lock - asynchronous execution of place bids and coordination for close auction • …… 6
17. Optimising Concurrency Control • Concurrency vs Performance • Concurrency • Different types of concurrency control options available • Different granularity • Performance • Throughput • Latency • Computational overhead 7
18. State of the art - Concurrency Control Synthesis • Hamsaz • Farzin Houshmand and Mohsen Lesani. Hamsaz: replication coordination analysis and synthesis. POPL, 2019 • Commutativity lattice • Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11 • Olisipo • Cheng Li, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18 • Callas • Chao Xie, Chunzhi Su, Cody Littley, Lorenzo Alvisi, Manos Kapritsos, and Yang Wang. 2015. High-performance ACID via modular concurrency control. SOSP ’15 • IPA • Valter Balegas, Sérgio Duarte, Carla Ferreira, Rodrigo Rodrigues, and Nuno Preguiça. IPA: invariant-preserving applications for weakly consistent replicated databases. VLDB ‘18 8
19. State of the art - Concurrency Control Synthesis • Hamsaz • Farzin Houshmand and Mohsen Lesani. Hamsaz: replication coordination analysis and synthesis. POPL, 2019 • Commutativity lattice • Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11 • Olisipo • Cheng Li, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18 • Callas • Chao Xie, Chunzhi Su, Cody Littley, Lorenzo Alvisi, Manos Kapritsos, and Yang Wang. 2015. High-performance ACID via modular concurrency control. SOSP ’15 • IPA • Valter Balegas, Sérgio Duarte, Carla Ferreira, Rodrigo Rodrigues, and Nuno Preguiça. IPA: invariant-preserving applications for weakly consistent replicated databases. VLDB ‘18 8 Conflict Graph Graph algorithms to parametrise protocols Middleware (no code
20. State of the art - Concurrency Control Synthesis • Hamsaz • Farzin Houshmand and Mohsen Lesani. Hamsaz: replication coordination analysis and synthesis. POPL, 2019 • Commutativity lattice • Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11 • Olisipo • Cheng Li, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18 • Callas • Chao Xie, Chunzhi Su, Cody Littley, Lorenzo Alvisi, Manos Kapritsos, and Yang Wang. 2015. High-performance ACID via modular concurrency control. SOSP ’15 • IPA • Valter Balegas, Sérgio Duarte, Carla Ferreira, Rodrigo Rodrigues, and Nuno Preguiça. IPA: invariant-preserving applications for weakly consistent replicated databases. VLDB ‘18 8 Conflict Graph Graph algorithms to parametrise protocols Middleware (no code Concurrency vs performance modeled as a lattice
21. State of the art - Concurrency Control Synthesis • Hamsaz • Farzin Houshmand and Mohsen Lesani. Hamsaz: replication coordination analysis and synthesis. POPL, 2019 • Commutativity lattice • Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11 • Olisipo • Cheng Li, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18 • Callas • Chao Xie, Chunzhi Su, Cody Littley, Lorenzo Alvisi, Manos Kapritsos, and Yang Wang. 2015. High-performance ACID via modular concurrency control. SOSP ’15 • IPA • Valter Balegas, Sérgio Duarte, Carla Ferreira, Rodrigo Rodrigues, and Nuno Preguiça. IPA: invariant-preserving applications for weakly consistent replicated databases. VLDB ‘18 8 Conflict Graph Graph algorithms to parametrise protocols Middleware (no code Concurrency vs performance modeled as a lattice Effect of workload on protocol selection
22. State of the art - Concurrency Control Synthesis • Hamsaz • Farzin Houshmand and Mohsen Lesani. Hamsaz: replication coordination analysis and synthesis. POPL, 2019 • Commutativity lattice • Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11 • Olisipo • Cheng Li, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18 • Callas • Chao Xie, Chunzhi Su, Cody Littley, Lorenzo Alvisi, Manos Kapritsos, and Yang Wang. 2015. High-performance ACID via modular concurrency control. SOSP ’15 • IPA • Valter Balegas, Sérgio Duarte, Carla Ferreira, Rodrigo Rodrigues, and Nuno Preguiça. IPA: invariant-preserving applications for weakly consistent replicated databases. VLDB ‘18 8 Conflict Graph Graph algorithms to parametrise protocols Middleware (no code Concurrency vs performance modeled as a lattice Effect of workload on protocol selection Federating hierarchical concurrency control mechanisms Different types of concurrency control
23. State of the art - Concurrency Control Synthesis • Hamsaz • Farzin Houshmand and Mohsen Lesani. Hamsaz: replication coordination analysis and synthesis. POPL, 2019 • Commutativity lattice • Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11 • Olisipo • Cheng Li, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18 • Callas • Chao Xie, Chunzhi Su, Cody Littley, Lorenzo Alvisi, Manos Kapritsos, and Yang Wang. 2015. High-performance ACID via modular concurrency control. SOSP ’15 • IPA • Valter Balegas, Sérgio Duarte, Carla Ferreira, Rodrigo Rodrigues, and Nuno Preguiça. IPA: invariant-preserving applications for weakly consistent replicated databases. VLDB ‘18 8 Conflict Graph Graph algorithms to parametrise protocols Middleware (no code Concurrency vs performance modeled as a lattice Effect of workload on protocol selection Federating hierarchical concurrency control mechanisms Synthesising coordination free data types Different types of concurrency control
24. • Lock coarsening • Pedro C. Diniz and Martin C. Rinard. Lock coarsening: Eliminating lock overhead in automatically parallelized object-based programs. Journal of Parallel and Distributed Computing 1998 • Dynamic feedback • Pedro C. Diniz and Martin C. Rinard. Dynamic feedback: An effective technique for adaptive computing. SIGPLAN 1997. • Performance prediction for coarse-grained locking • Vitaly Aksenov, Dan Alistarh, and Petr Kuznetsov. Brief announcement: Performance prediction for coarse-grained locking. PODC ’18, 9 State of the art - Lock granularity
25. • Lock coarsening • Pedro C. Diniz and Martin C. Rinard. Lock coarsening: Eliminating lock overhead in automatically parallelized object-based programs. Journal of Parallel and Distributed Computing 1998 • Dynamic feedback • Pedro C. Diniz and Martin C. Rinard. Dynamic feedback: An effective technique for adaptive computing. SIGPLAN 1997. • Performance prediction for coarse-grained locking • Vitaly Aksenov, Dan Alistarh, and Petr Kuznetsov. Brief announcement: Performance prediction for coarse-grained locking. PODC ’18, 9 State of the art - Lock granularity Systematic coarsening of locks with policies Granularity vs performance trade-off
26. • Lock coarsening • Pedro C. Diniz and Martin C. Rinard. Lock coarsening: Eliminating lock overhead in automatically parallelized object-based programs. Journal of Parallel and Distributed Computing 1998 • Dynamic feedback • Pedro C. Diniz and Martin C. Rinard. Dynamic feedback: An effective technique for adaptive computing. SIGPLAN 1997. • Performance prediction for coarse-grained locking • Vitaly Aksenov, Dan Alistarh, and Petr Kuznetsov. Brief announcement: Performance prediction for coarse-grained locking. PODC ’18, 9 State of the art - Lock granularity Systematic coarsening of locks with policies Choosing coarsening policy depending on the dynamic workload Granularity vs performance trade-off
27. • Lock coarsening • Pedro C. Diniz and Martin C. Rinard. Lock coarsening: Eliminating lock overhead in automatically parallelized object-based programs. Journal of Parallel and Distributed Computing 1998 • Dynamic feedback • Pedro C. Diniz and Martin C. Rinard. Dynamic feedback: An effective technique for adaptive computing. SIGPLAN 1997. • Performance prediction for coarse-grained locking • Vitaly Aksenov, Dan Alistarh, and Petr Kuznetsov. Brief announcement: Performance prediction for coarse-grained locking. PODC ’18, 9 State of the art - Lock granularity Systematic coarsening of locks with policies Choosing coarsening policy depending on the dynamic workload Modelling a lock to predict performance Granularity vs performance trade-off
28. Overview 10 Specification & Invariants Conflicting Methods & Synchronisation Conditions Concurrency Control Synthesis Optimised Concurrency Control Mechanism Static Analysis Static Workload
29. Stage 1 : Static Analysis 11 Specification & Invariants Conflicting Methods & Synchronisation Conditions Concurrency Control Synthesis + Static Workload Optimised Concurrency Control Mechanism Static Analysis Specification & Invariants Conflicting Methods Synchronisation condition generation Synchronisation Conditions Corrected specification Verification
30. Auction : Static Analysis Conflicts detected • Add product - unregister seller • Create auction - unregister seller • Add to lot - add to lot • Add to lot - start auction • Add to lot - remove auction • Add to lot - remove product • Start auction - remove from lot • Place bid - close auction • Place bid - unregister buyer • Close auction - remove bid 12
31. Auction : Synchronisation Condition Generation 13 • Add product - unregister seller seller • Create auction - unregister seller seller • Add to lot - add to lot seller, (product, seller) • Add to lot - start auction auction • Add to lot - remove auction auction • Add to lot - remove product (product, seller) • Start auction - remove from lot auction • Place bid - close auction auction • Place bid - unregister buyer buyer • Close auction - remove bid auction
32. Stage 2 : Concurrency Control Synthesis 14 Specification & Invariants Static Analysis Conflicting Methods & Synchronisation Conditions Concurrency Control Synthesis + Static Workload Optimised Concurrency Control Mechanism Conflicting Methods & Synchronisation Conditions Preprocessing Weighted subgraphs Concurrency control type Refined granularity for synchronisation conditions Optimised concurrency control Performance Evaluation Concurrency Control Type Selection Coarsening Static Workload Characteristics
33. Conflicting Methods & Synchronisation Conditions Static Workload Characteristics Grouping Weighted subgraphs Stage 2.1 : Preprocessing 15 Groups of methods Weight addition Filtering Groups of dependent methods
34. Auction : Grouping 16 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid Register seller Register buyer Add to lot Start auction Remove from lot Remove auction Remove product
35. Auction : Filtering 17 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid Register seller Register buyer Add to lot Start auction Remove from lot Remove auction Remove product
36. Auction : Filtering 17 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid Add to lot Start auction Remove from lot Remove auction Remove product
37. Auction : Weight addition 18 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid 10 5 500 5 200 1 50 Add to lot Start auction Remove from lot Remove auction Remove product 20 5 2 5 5
38. Stage 2.2 : Coarsening 19 Weighted subgraphs of dependent methods with synchronisation conditions Data Coarsener Coarsened synchronisation conditions Computational Coarsener + Policy
39. Stage 2.2 : Coarsening 19 Weighted subgraphs of dependent methods with synchronisation conditions Data Coarsener Coarsened synchronisation conditions Computational Coarsener + Policy Coarsening synchronisation conditions Reducing locks to be acquired by a single operation
40. Stage 2.2 : Coarsening 19 Weighted subgraphs of dependent methods with synchronisation conditions Data Coarsener Coarsened synchronisation conditions Computational Coarsener + Policy Coarsening synchronisation conditions Reducing locks to be acquired by a single operation 1. No coarsening 2. Synchronisation condition-based coarsening 3. Group-based coarsening
41. Auction : Data Coarsening 20 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid seller seller auction auction buyer Add to lot Start auction Remove from lot Remove auction Remove product (product, seller) seller, (product, seller) auction auction auction
42. Auction : Data Coarsening 20 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid seller seller auction auction buyer Add to lot Start auction Remove from lot Remove auction Remove product auction auction auction seller seller
43. Auction : No coarsening 21 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid seller seller auction auction buyer Add to lot Start auction Remove from lot Remove auction Remove product seller seller auction auction auction
44. Auction : Synchronisation condition-based coarsening 22 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid seller auction buyer Add to lot Start auction Remove from lot Remove auction Remove product seller auction
45. Auction : Group-based coarsening 23 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid seller (auction, buyer) Add to lot Start auction Remove from lot Remove auction Remove product (seller, auction)
46. Stage 2.3 : Concurrency Control Type Selection 24 Concurrency Control Type Selection Concurrency Control Type Weighted subgraphs of dependent methods with synchronisation conditions
47. Stage 2.3 : Concurrency Control Type Selection 24 Concurrency Control Type Selection Concurrency Control Type Weighted subgraphs of dependent methods with synchronisation conditions 1. Mutex • Equal weights 2. Async/wait lock • Unequal weights, async to heavier operation
48. Auction : No coarsening 25 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid seller seller auction auction buyer 10 5 500 5 200 1 50 Add to lot Start auction Remove from lot Remove auction Remove product seller seller auction auction auction 20 5 2 2 5 Async Wait Async Wait Async Wait Async Wait Mutex Async Wait A syn c W ait Async Wait Async Wait Async Wait
49. Auction : Synchronisation condition-based coarsening 26 Place bid Close auction Unregister buyer Remove bid auction buyer 500 200 50 5 Async Wait Wait Async Async Add product Unregister seller Create auction seller 10 5 1 Wait Async Async Add to lot Start auction Remove from lot Remove auction Remove product seller auction 20 5 2 2 5 Wait Async Async Wait Wait Async
50. Auction : Group-based coarsening 27 Add product Unregister seller Create auction Place bid Close auction Unregister buyer Remove bid seller (auction, buyer) Add to lot Start auction Remove from lot Remove auction Remove product (seller, auction) 10 5 500 200 1 50 5 20 5 2 2 5 Wait Async Async Wait Async Async Wait Wait Wait Async Async Async
51. Future work • Year 2: • Validation of methodology • Revisiting synchronisation conditions and iterating • Year 3: • Considering workload phases and transitions 28
52. Summary 29 Optimising Concurrency Control Year 1 Year 2 Year 3 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Literature review Focused on static analysis Focused on optimising concurrency control Focused on dynamic workload Static analysis • Published at PaPoC ’19 • Submitted to OOPSLA ‘19 Methodology development Formulating methodology Iterative validation Improvising • Numerical constraints • Dynamic coarsening policy • Different types of concurrency control Workload phases and transitions Categorize workloads into phases Study the effect of transitions Done In progress Planned
53. State of the Art •Farzin Houshmand and Mohsen Lesani. 2019. Hamsaz: replication coordination analysis and synthesis. POPL, 2019 •Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11 •Diniz, Pedro C., and Martin C. Rinard. Lock coarsening: Eliminating lock overhead in automatically parallelized object- based programs. Journal of Parallel and Distributed Computing 49, no. 2 (1998): 218-244. •Tom Magrino, Jed Liu, Nate Foster, Johannes Gehrke, and Andrew C. Myers. Efficient, Consistent Distributed Computation with Predictive Treaties. EuroSys '19 •Marcos K. Aguilera, Tudor David, Rachid Guerraoui, and Junxiong Wang. Locking Timestamps versus Locking Objects. PODC ’18 •Li, Cheng, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18 30
54. State of the Art contd.. • Matthew Milano and Andrew C. Myers. MixT: a language for mixing consistency in geodistributed transactions. PLDI 2018 • Zhang, Yang, Russell Power, Siyuan Zhou, Yair Sovran, Marcos K. Aguilera, and Jinyang Li. Transaction chains: achieving serializability with low latency in geo-distributed storage systems. SOSP 2013. • Shuai Mu, Sebastian Angel, and Dennis Shasha. Deferred Runtime Pipelining for contentious multicore software transactions. EuroSys '19 • Gowtham Kaki, Kapil Earanky, KC Sivaramakrishnan, and Suresh Jagannathan. Safe replication through bounded concurrency verification. OOPSLA 2018 • Roy, S., Kot, L., Bender, G., Ding, B., Hojjat, H., Koch, C., Foster, N., and Gehrke, J. The homeostasis protocol: Avoiding transaction coordination through program analysis. SIGMOD 2015. • Liu, J., Magrino, T., Arden, O., George, M. D., and Myers, A. C. Warranties for faster strong consistency. NSDI 2014 31
55. • Hamsaz • + presents conflict information as a graph • + presents two protocols, automatically generated • + application code need not be modified, conflict handling is performed by a middleware using the two protocols • - programmer needs to select any one protocol • - does not distinguish the conditions under which the conflicting operations must synchronise or can run asynchronously • - doesn’t discuss granularity • - workload dynamics are not considered 32 State of the art Farzin Houshmand and Mohsen Lesani. Hamsaz: replication coordination analysis and synthesis. POPL, 2019
56. • Indigo • Violation-avoidance or invariant repair options • Explicit consistency • code is instrumented • workload dynamics are not considered 33 State of the art Valter Balegas, Sérgio Duarte, Carla Ferreira, Rodrigo Rodrigues, Nuno Preguiça, Mahsa Najafzadeh, and Marc Shapiro. Putting consistency back into eventual consistency. EuroSys ’15
57. • Commutativity lattice • + 4 protocols proposed as a lattice, the higher, the more parallel • + systematic traversal through the lattice • - meant for parallel programs, not for replicated states • - workload is not considered 34 State of the art Milind Kulkarni, Donald Nguyen, Dimitrios Prountzos, Xin Sui, and Keshav Pingali. Exploiting the commutativity lattice. PLDI '11
58. • Olisipo • 2 protocols resembling mutex and r/w locks - Sym and Asym • - Workload is considered as a parameter, but not parametrized • + partially ordered consistency • - no fine grained with method parameter • - just broadcast protocols, without considering specific fine-grained coordination requirements 35 State of the art Li, Cheng, Nuno Preguica, and Rodrigo Rodrigues. Fine-grained consistency for geo-replicated systems. USENIX ATC 18
59. • Callas • Looks into transaction isolation properties for sharded key-value stores • + hierarchical modular concurrency control • + modular and extensible • + supports diff concurrency control schemes - single version to multi version and lock-based to timestamp-based • + considers workloads • + transactions are partitioned in groups and diff concurrency control mechanism allowed per group • - isolation is the correctness property, dealing with databases, not distributed apps • - not dealing with replication 36 State of the art Chao Xie, Chunzhi Su, Cody Littley, Lorenzo Alvisi, Manos Kapritsos, and Yang Wang. 2015. High-performance ACID via modular concurrency control. SOSP ’15
60. PIF Activities Project meetings • RainbowFS workshop • 20/05/2018 - Barsac • 04/03/2019 - Annecy • Lightkone workshop • 12/06/2018 - Maastricht • 12/11/2018 - Paris • 28/01/2019 - Barcelona • 14/05/2019 - Kaiserslautern • 13/06/2019 - Munich 37 Conferences • 25/03/2019 - EuroSys + PaPoC 2019 • 15/03/2019 - IRIF Verification Seminar Paper submissions • ESOP 2019 - rejected • PaPoC 2019 - accepted • OOPSLA 2019 - under review Non-scientific • Français Langue Etranger • Rules for publishing a scientific article