noria
play

Noria Dynamic, partially-stateful data-flow for high-performance - PowerPoint PPT Presentation

Nov 07, 2022 •499 likes •1.6k views

Noria Dynamic, partially-stateful data-flow for high-performance Web applications Jon Gjengset Malte Schwarzkopf Jonathan Behrens Lara Timb Arajo Martin Ek Eddie Kohler M. Frans Kaashoek Robert Morris 2 Frontend 2

  1. Upquery responses in total order with updates Goal: upquery restores state as if present all along. � 17

  2. Upquery responses in total order with updates Goal: upquery restores state as if present all along. 3 2 resulting state 3 respects total order 2 1 � 17

  3. Upquery responses in total order with updates Goal: upquery restores state as if present all along. 3 2 2 resulting state 3 3 resulting state 2 respects total order 2 violates total order 2 1 1 � 17

  4. Upquery responses in total order with updates Goal: upquery restores state as if present all along. 3 2 2 resulting state 3 3 resulting state 2 respects total order 2 violates total order 2 1 1 More complex cases: merged upquery responses, evictions ( Paper ). � 17

  5. Challenges implementing partially-stateful data-flow 1. Concurrent upqueries and forward processing — races! Must maintain correctness under concurrency! 2. Update processing may require absent state � 18

  6. Challenges implementing partially-stateful data-flow 1. Concurrent upqueries and forward processing — races! Must maintain correctness under concurrency! 2. Update processing may require absent state 2 COUNT absent 3 … � 18

  7. Challenges implementing partially-stateful data-flow 1. Concurrent upqueries and forward processing — races! Must maintain correctness under concurrency! 2. Update processing may require absent state Drop updates that touch absent 2 state, future upquery repeats them. COUNT absent 3 … � 18

  8. Challenges implementing partially-stateful data-flow 1. Concurrent upqueries and forward processing — races! Must maintain correctness under concurrency! (see Paper) 2. Update processing may require absent state Drop updates that touch absent 2 state, future upquery repeats them. COUNT absent 3 … � 18

  9. Noria implementation � 19

  10. Noria implementation � 19

  11. Noria implementation MySQL adapter � 19

  12. Noria implementation Data-flow graph Transform MySQL adapter � 19

  13. Noria implementation Data-flow graph Transform MySQL adapter � 19

  14. Noria implementation Data-flow graph Transform • 45k lines of Rust + 15k libraries • RocksDB for base table storage MySQL adapter • ZooKeeper for leader election � 19

  15. Evaluation 1. Can Noria improve a real web application’s performance? 2. How does Noria compare to alternative approaches? 3. Can Noria change queries without downtime? � 20

  16. Evaluation 1. Can Noria improve a real web application’s performance? 2. How does Noria compare to alternative approaches? 3. Can Noria change queries without downtime? Setup Amazon EC2 c5.4xlarge instance (16 vCPUs) Open-loop clients, measuring latency & throughput � 20

  17. Evaluation 1. Can Noria improve a real web application’s performance? 2. How does Noria compare to alternative approaches? 3. Can Noria change queries without downtime? Setup Amazon EC2 c5.4xlarge instance (16 vCPUs) Open-loop clients, measuring latency & throughput comparison with di ff erential dataflow } see Paper multi-machine experiments 
 � 20

  18. Case study: Lobsters (http://lobste.rs) � 21

  19. Case study: Lobsters (http://lobste.rs) ‣ Ruby-on-Rails application 
 with MySQL backend � 21

  20. Case study: Lobsters (http://lobste.rs) ‣ Ruby-on-Rails application 
 with MySQL backend ‣ Hand-optimized by developers to pre-compute aggregations � 21

  21. Case study: Lobsters (http://lobste.rs) ‣ Ruby-on-Rails application 
 with MySQL backend ‣ Hand-optimized by developers to pre-compute aggregations ‣ Noria data-flow with 
 235 operators, 35 views � 21

  22. Case study: Lobsters (http://lobste.rs) ‣ Ruby-on-Rails application 
 with MySQL backend ‣ Hand-optimized by developers to pre-compute aggregations ‣ Noria data-flow with 
 235 operators, 35 views ‣ Emulate production load � 21

  23. Can Noria improve Lobsters’ performance? � 22

  24. Can Noria improve Lobsters’ performance? Better Better � 22

  25. Can Noria improve Lobsters’ performance? Better Better � 22

  26. Can Noria improve Lobsters’ performance? Better Better � 22

  27. Can Noria improve Lobsters’ performance? Better Better Noria with natural queries supports 5x MySQL’s throughput. � 22

  28. How does Noria compare to alternatives? Better Better � 23

  29. How does Noria compare to alternatives? Better Better ‣ Zipf-distributed story ID, 
 95% reads, 5% writes ‣ No TX, all in-memory � 23

  30. How does Noria compare to alternatives? Better Better ‣ Zipf-distributed story ID, 
 95% reads, 5% writes ‣ No TX, all in-memory � 23

  31. How does Noria compare to alternatives? Better Better ‣ Zipf-distributed story ID, 
 95% reads, 5% writes ‣ No TX, all in-memory � 24

  32. How does Noria compare to alternatives? Better Better ‣ Zipf-distributed story ID, 
 95% reads, 5% writes ‣ No TX, all in-memory � 24

  33. How does Noria compare to alternatives? Better Better ‣ Zipf-distributed story ID, 
 Noria outperforms an in-memory key- 95% reads, 5% writes value store and simplifies its interface. ‣ No TX, all in-memory � 24

  34. Can Noria change queries without downtime? Stories Votes COUNT JOIN FILTER 3 2 � 25

  35. Can Noria change queries without downtime? Ratings Stories Votes ⭐⭐⭐ ⭐ COUNT ⭐⭐ JOIN AVG COUNT FILTER 3 JOIN 3 ⭐ 2 1.5 ⭐ � 25

  36. Can Noria change queries without downtime? Better � 26

  37. Can Noria change queries without downtime? Better new table & query added � 26

More recommend