efficient hardware assisted logging
play

Efficient Hardware-assisted Logging with Asynchronous and Direct - PowerPoint PPT Presentation

Mar 10, 2024 •383 likes •1.95k views

Efficient Hardware-assisted Logging with Asynchronous and Direct Update for Persistent Memory Jungi Jeong , Chang Hyun Park, Jaehyuk Huh, and Seungryoul Maeng International Symposium on Microarchitecture (MICRO) 2018 Storage-Class Memory

  1. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store B-> next= C store B-> next= C store B-> prev= A store B-> prev= A store A-> next= B store A-> next= B store C-> prev= B store C-> prev= B cache-flush 2018-10-23 48

  2. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store B-> next= C store B-> next= C Log store B-> prev= A store B-> prev= A store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B cache-flush 2018-10-23 49

  3. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store B-> next= C store B-> next= C Log store B-> prev= A store B-> prev= A store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B cache-flush Data Update 2018-10-23 50

  4. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store log[0]= A-> next store B-> next= C store B-> next= C Log store log[1]= C-> prev store B-> prev= A store B-> prev= A cache-flush store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B sfence cache-flush store B-> next= C store B-> prev= A Data store A-> next= B Update store C-> prev= B cache-flush sfence 2018-10-23 51

  5. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store log[0]= A-> next store B-> next= C store B-> next= C Log store log[1]= C-> prev store B-> prev= A store B-> prev= A cache-flush store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B sfence cache-flush store B-> next= C store B-> prev= A Data store A-> next= B Update store C-> prev= B cache-flush sfence 2018-10-23 52

  6. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store log[0]= A-> next store B-> next= C store B-> next= C Log store log[1]= C-> prev store B-> prev= A store B-> prev= A cache-flush store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B sfence cache-flush store B-> next= C store B-> prev= A Data store A-> next= B Update store C-> prev= B cache-flush sfence Persist-ordering with store-fence 2018-10-23 53

  7. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store log[0]= A-> next store B-> next= C store B-> next= C Log store log[1]= C-> prev store B-> prev= A store B-> prev= A cache-flush store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B sfence cache-flush store B-> next= C store B-> prev= A Data store A-> next= B Update store C-> prev= B cache-flush sfence Persist-ordering with store-fence 2018-10-23 54

  8. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store log[0]= A-> next store B-> next= C store B-> next= C Log store log[1]= C-> prev store B-> prev= A store B-> prev= A cache-flush store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B sfence cache-flush store B-> next= C store B-> prev= A Data store A-> next= B Update store C-> prev= B cache-flush sfence Persist-ordering with store-fence 2018-10-23 55

  9. Atomic Durability through Logging • Transaction • All stores in a transaction become durable all together or nothing • Ex) Atomic durability in software Durability with Atomicity and ordering with cache-flush write-ahead logging store log[0]= A-> next store B-> next= C store B-> next= C Log store log[1]= C-> prev store B-> prev= A store B-> prev= A cache-flush store A-> next= B store A-> next= B Write store C-> prev= B store C-> prev= B sfence cache-flush store B-> next= C store B-> prev= A Data store A-> next= B Update store C-> prev= B cache-flush sfence Persist-ordering with store-fence 2018-10-23 56

  10. HW-assisted Logging 57 2018-10-23

  11. HW-assisted Logging • Simple programming model • HW is responsible for 1) log-write and 2) data update • Advantages over software-logging • Fine-grained ordering & less CPU cycles 58 2018-10-23

  12. HW-assisted Logging • Simple programming model Transa sact ct ion_begin( ) ( ) store B-> next= C … store C-> prev= B Transact ion_end( ) ( ) • HW is responsible for 1) log-write and 2) data update • Advantages over software-logging • Fine-grained ordering & less CPU cycles 59 2018-10-23

  13. HW-assisted Logging • Simple programming model Transa sact ct ion_begin( ) ( ) store B-> next= C store B-> next= C … … store C-> prev= B store C-> prev= B Transact ion_end( ) ( ) • HW is responsible for 1) log-write and 2) data update • Advantages over software-logging • Fine-grained ordering & less CPU cycles 60 2018-10-23

  14. HW-assisted Logging • Simple programming model Transa sact ct ion_begin( ) ( ) store B-> next= C store B-> next= C … … store C-> prev= B store C-> prev= B Transact ion_end( ) ( ) • HW is responsible for 1) log-write and 2) data update • Advantages over software-logging • Fine-grained ordering & less CPU cycles 61 2018-10-23

  15. HW-assisted Logging • Simple programming model Transa sact ct ion_begin( ) ( ) store B-> next= C store B-> next= C … … store C-> prev= B store C-> prev= B Transact ion_end( ) ( ) • HW is responsible for 1) log-write and 2) data update Caches Log Ctrl. Processor 2) Data-Update 1) Log-Write NVM Data NVM Log NVM • Advantages over software-logging • Fine-grained ordering & less CPU cycles 62 2018-10-23

  16. HW-assisted Logging • Simple programming model Transa sact ct ion_begin( ) ( ) store B-> next= C store B-> next= C … … store C-> prev= B store C-> prev= B Transact ion_end( ) ( ) • HW is responsible for 1) log-write and 2) data update Caches Log Ctrl. Processor 2) Data-Update 1) Log-Write NVM Data NVM Log NVM • Advantages over software-logging • Fine-grained ordering & less CPU cycles Fence Log B Store B Log B Store B Log A Store A Log A Store A (a) Ordering with SW (b) Ordering with HW 63 2018-10-23

  17. Past Proposal: Undo-based HW-Logging Processor Caches Log Ctrl. NVM Data NVM Log NVM • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 64 2018-10-23

  18. Past Proposal: Undo-based HW-Logging • Store old value in logs • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. NVM Data NVM Log NVM • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 65 2018-10-23

  19. Past Proposal: Undo-based HW-Logging • Store old value in logs • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. 1) Store old value in NVM logs NVM Data NVM Log NVM • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 66 2018-10-23

  20. Past Proposal: Undo-based HW-Logging • Store old value in logs • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. 1) Store old value in NVM logs NVM Data NVM Log NVM • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 67 2018-10-23

  21. Past Proposal: Undo-based HW-Logging • Store old value in logs Addr • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. 1) Store old value in NVM logs NVM Data NVM Log NVM • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 68 2018-10-23

  22. Past Proposal: Undo-based HW-Logging • Store old value in logs Old Value Addr • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. 2) Update data 1) Store old value in NVM in NVM logs NVM Data NVM Log NVM • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 69 2018-10-23

  23. Past Proposal: Undo-based HW-Logging • Store old value in logs Old Value Addr • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. 2) Update data 1) Store old value in NVM in NVM logs NVM Data NVM Log NVM Long critical path due to synchronous data-update • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 70 2018-10-23

  24. Past Proposal: Undo-based HW-Logging • Store old value in logs Old Value Addr • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. 2) Update data 1) Store old value in NVM in NVM logs NVM Data NVM Log NVM Long critical path due to synchronous data-update • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 71 2018-10-23

  25. Past Proposal: Undo-based HW-Logging • Store old value in logs Old Value Addr • Update data in NVM before commit  Synchronous data-update Processor Caches Log Ctrl. 2) Update data 1) Store old value in NVM in NVM logs NVM Data NVM Log NVM Long critical path due to synchronous data-update • A. Joshi et al. HPCA 2017. • S. Shin et al. ISCA 2017. 72 2018-10-23

  26. Past Proposal: Redo-based HW-Logging Addr New Value Processor Caches Log Ctrl. NVM NVM Data NVM Log • K. Doshi et al. HPCA 2016. 73 2018-10-23

  27. Past Proposal: Redo-based HW-Logging • Store new value in logs Addr New Value • Update data in NVM after commit  Asynchronous data-update • However, update by reading log entries from NVM  I ndirect data-update Processor Caches Log Ctrl. NVM NVM Data NVM Log • K. Doshi et al. HPCA 2016. 74 2018-10-23

  28. Past Proposal: Redo-based HW-Logging • Store new value in logs Addr New Value • Update data in NVM after commit  Asynchronous data-update • However, update by reading log entries from NVM  I ndirect data-update Processor Caches Log Ctrl. 1) Store new value in NVM logs NVM NVM Data NVM Log • K. Doshi et al. HPCA 2016. 75 2018-10-23

  29. Past Proposal: Redo-based HW-Logging • Store new value in logs Addr New Value • Update data in NVM after commit  Asynchronous data-update • However, update by reading log entries from NVM  I ndirect data-update Processor Caches Log Ctrl. 1) Store new value in NVM logs NVM NVM Data NVM Log • K. Doshi et al. HPCA 2016. 76 2018-10-23

  30. Past Proposal: Redo-based HW-Logging • Store new value in logs Addr New Value • Update data in NVM after commit  Asynchronous data-update • However, update by reading log entries from NVM  I ndirect data-update Processor Caches Log Ctrl. 1) Store new value 2) Update data in NVM logs in NVM NVM NVM Data NVM Log • K. Doshi et al. HPCA 2016. 77 2018-10-23

  31. Past Proposal: Redo-based HW-Logging • Store new value in logs Addr New Value • Update data in NVM after commit  Asynchronous data-update • However, update by reading log entries from NVM  I ndirect data-update Processor Caches Log Ctrl. 1) Store new value 2) Update data in NVM logs in NVM NVM NVM Data NVM Log • K. Doshi et al. HPCA 2016. 78 2018-10-23

  32. Past Proposal: Redo-based HW-Logging • Store new value in logs Addr New Value • Update data in NVM after commit  Asynchronous data-update • However, update by reading log entries from NVM  I ndirect data-update Processor Caches Log Ctrl. 1) Store new value 2) Update data in NVM logs in NVM NVM NVM Data NVM Log Wastes extra NVM bandwidth for reading logs from NVM • K. Doshi et al. HPCA 2016. 79 2018-10-23

  33. Past Proposal: Undo-Redo HW-Logging Addr Old Value New Value Processor Caches Log Buffer NVM NVM Data NVM Log • M. Ogleari et al. HPCA 2018. 80 2018-10-23

  34. Past Proposal: Undo-Redo HW-Logging • Store both old and new value in logs  Larger log sizes Addr Old Value New Value • Update data in NVM after commit Processor Caches Log Buffer NVM NVM Data NVM Log • M. Ogleari et al. HPCA 2018. 81 2018-10-23

  35. Past Proposal: Undo-Redo HW-Logging • Store both old and new value in logs  Larger log sizes Addr Old Value New Value • Update data in NVM after commit Processor Caches Log Buffer 1) Store both old and new value in NVM logs NVM NVM Data NVM Log Requires more NVM writes for storing logs in NVM • M. Ogleari et al. HPCA 2018. 82 2018-10-23

  36. Past Proposal: Undo-Redo HW-Logging • Store both old and new value in logs  Larger log sizes Addr Old Value New Value • Update data in NVM after commit Processor Caches Log Buffer 1) Store both old 2) Update data and new value in NVM in NVM logs NVM NVM Data NVM Log Requires more NVM writes for storing logs in NVM • M. Ogleari et al. HPCA 2018. 83 2018-10-23

  37. Past Proposal: Undo-Redo HW-Logging • Store both old and new value in logs  Larger log sizes Addr Old Value New Value • Update data in NVM after commit Processor Caches Log Buffer 1) Store both old 2) Update data and new value in NVM in NVM logs NVM NVM Data NVM Log Requires more NVM writes for storing logs in NVM • M. Ogleari et al. HPCA 2018. 84 2018-10-23

  38. Past Proposal: Undo-Redo HW-Logging • Store both old and new value in logs  Larger log sizes Addr Old Value New Value • Update data in NVM after commit Processor Caches Log Buffer 1) Store both old 2) Update data and new value in NVM in NVM logs NVM NVM Data NVM Log Requires more NVM writes for storing logs in NVM • M. Ogleari et al. HPCA 2018. 85 2018-10-23

  39. Past Proposal: Undo-Redo HW-Logging • Store both old and new value in logs  Larger log sizes Addr Old Value New Value • Update data in NVM after commit Processor Caches Log Buffer 1) Store both old 2) Update data and new value in NVM in NVM logs NVM NVM Data NVM Log Requires more NVM writes for storing logs in NVM • M. Ogleari et al. HPCA 2018. 86 2018-10-23

  40. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous 87 2018-10-23

  41. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous 1.4 1.2 1.2 1 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Workloads 88 2018-10-23

  42. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Workloads 89 2018-10-23

  43. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 90 2018-10-23

  44. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 91 2018-10-23

  45. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 92 2018-10-23

  46. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 93 2018-10-23

  47. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 94 2018-10-23

  48. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 95 2018-10-23

  49. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 96 2018-10-23

  50. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 97 2018-10-23

  51. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 98 2018-10-23

  52. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 99 2018-10-23

  53. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 100 2018-10-23

  54. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better Lower is Better 1 0.8 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 101 2018-10-23

  55. Past Proposals: Summary Log-Write Data-Update Drawback Synchronous ATOM [HPCA 2017] Undo Direct Long Critical Path Proteus [ISCA 2017] Undo Direct Synchronous I ndirect Waste NVM Bandwidth Wrap [HPCA 2016] Redo Asynchronous UndoRedo More Log Write FWB [HPCA 2018] Direct Asynchronous Undo-Friendly Redo-Friendly Cycles per Transaction (CPT) Cycles per Transaction (CPT) 1.4 1.2 1.2 1 Lower is Better Lower is Better 1 0.8 Trade-offs exist! 0.8 0.6 0.6 0.4 0.4 1 2 3 1 2 3 Large & Sequential Small & Random Workloads Workloads 102 2018-10-23

  56. Design Goal & Challenges 103 2018-10-23

  57. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches 104 2018-10-23

  58. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Caches 105 2018-10-23

  59. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Processor Caches NVM Log 106 2018-10-23

  60. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Processor Caches NVM Log 107 2018-10-23

  61. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Processor Caches NVM Log 108 2018-10-23

  62. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Processor Caches NVM Log NVM Data 109 2018-10-23

  63. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Processor Caches NVM Log NVM Data NVM 110 2018-10-23

  64. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Processor Caches NVM Log NVM Data NVM 111 2018-10-23

  65. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Caches NVM Log NVM Data NVM 112 2018-10-23

  66. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Caches NVM Log NVM Data NVM 113 2018-10-23

  67. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Tx3 Caches NVM Log NVM Data NVM 114 2018-10-23

  68. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Tx3 Caches NVM Log NVM Data Tx1 NVM 115 2018-10-23

  69. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Tx3 Caches NVM Log NVM Data Tx1 NVM Tx2 116 2018-10-23

  70. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Tx3 Caches NVM Log NVM Data Tx1 NVM Tx2 117 2018-10-23

  71. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Tx3 Caches NVM Log NVM Data Tx1 NVM Tx2 118 2018-10-23

  72. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Transaction # 1 Tx3 Caches NVM Log NVM Data Tx1 NVM Tx2 119 2018-10-23

  73. Design Goal & Challenges • Redo log with asynchronous & direct update to NVM • Challenge # 1: tracking write-sets of previous transactions • Without data update, logs keep growing • Challenge # 2: handling an early-eviction • Eviction of uncommitted changes from volatile CPU caches Tx1 Processor Tx2 Transaction # 1 (committed) Tx3 Caches NVM Log NVM Data Tx1 NVM Tx2 120 2018-10-23

Recommend

Efficient Hardware-based Undo+Redo Logging for Persistent Memory Systems Matheus Ogleari Prof.

Efficient Hardware-based Undo+Redo Logging for Persistent Memory Systems Matheus Ogleari Prof.

Efficient Hardware-based Undo+Redo Logging for Persistent Memory Systems Matheus Ogleari Prof. Ethan Miller Prof. Jishen Zhao 1 Overview Background/Motivation Hardware-Driven Logging Design Evaluation 2 Background

232 views • 20 slides
CCHL: Compression-Consolidation Hardware Logging for Efficient Failure-Atomic Persistent Memory

CCHL: Compression-Consolidation Hardware Logging for Efficient Failure-Atomic Persistent Memory

CCHL: Compression-Consolidation Hardware Logging for Efficient Failure-Atomic Persistent Memory Updates Xueliang Wei, Dan Feng, Wei Tong, Jingning Liu, Chengning Wang, Liuqing Ye Huazhong University of Science and Technology Persistent Memory

817 views • 38 slides
HOOP: Efficient Hardware-Assisted Out-of-Place Update for Non-Volatile Memory Miao Cai Chance

HOOP: Efficient Hardware-Assisted Out-of-Place Update for Non-Volatile Memory Miao Cai Chance

HOOP: Efficient Hardware-Assisted Out-of-Place Update for Non-Volatile Memory Miao Cai Chance Coats Jian Huang Systems Platform Research Group Non-Volatile Memory is a Revolutionary Technology Close-to-DRAM Performance Data

367 views • 21 slides
Efficient Data Structures for Tamper-Evident Logging Scott A. Crosby Dan S. Wallach Rice

Efficient Data Structures for Tamper-Evident Logging Scott A. Crosby Dan S. Wallach Rice

Efficient Data Structures for Tamper-Evident Logging Scott A. Crosby Dan S. Wallach Rice University Everyone has logs Tamper evident solutions Current commercial solutions Write only hardware appliances Security depends on

1.32k views • 82 slides
Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW)

Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW)

Machine Instructions Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW) Term 2 2020 1 Machine Instructions Where we are at In the last lecture we introduced efficient algorithms for critical

301 views • 18 slides
Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW)

Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW)

Machine Instructions Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW) Term 2 2020 1 Machine Instructions Where we are at In the last lecture we introduced efficient algorithms for critical

728 views • 11 slides
Practical Secure Two-Party Computation and Applications Lecture 4: Hardware-Assisted

Practical Secure Two-Party Computation and Applications Lecture 4: Hardware-Assisted

Practical Secure Two-Party Computation and Applications Lecture 4: Hardware-Assisted Cryptographic Protocols Estonian Winter School in Computer Science 2016 Motivation 2 Two Areas Cryptographic Protocols Secure Hardware strong security

759 views • 52 slides
Proteus: A Flexible and Fast Software supported Hardware Logging approach for NVM Seunghee Shin,

Proteus: A Flexible and Fast Software supported Hardware Logging approach for NVM Seunghee Shin,

Proteus: A Flexible and Fast Software supported Hardware Logging approach for NVM Seunghee Shin, Satish Tirukkovalluri, James Tuck, and Yan Solihin North Carolina State University The 2018 Non-Volatile Memories Workshop (NVMW 2018) 1

551 views • 17 slides
Logging and Recovery Module 6, Lectures 3 and 4 If you are going to be in the logging business,

Logging and Recovery Module 6, Lectures 3 and 4 If you are going to be in the logging business,

Logging and Recovery Module 6, Lectures 3 and 4 If you are going to be in the logging business, one of the things that you have to do is to learn about heavy equipment. Robert VanNatta, Logging History of Columbia County Database Management

534 views • 26 slides
Taming Transactions: Towards Hardware-Assisted Control Flow Integrity using Transactional Memory

Taming Transactions: Towards Hardware-Assisted Control Flow Integrity using Transactional Memory

The 19th International Symposium on Research in Attacks, Intrusions and Defenses (RAID 2016) Taming Transactions: Towards Hardware-Assisted Control Flow Integrity using Transactional Memory Marius Muench, Fabio Pagani, Yan Shoshitaishvili,

559 views • 25 slides
The Promises and Pitfalls of Hardware-Assisted Security Alexandra Dmitrienko

The Promises and Pitfalls of Hardware-Assisted Security Alexandra Dmitrienko

The Promises and Pitfalls of Hardware-Assisted Security Alexandra Dmitrienko Julius-Maximilians-Universitt Wrzburg alexandra.dmitrienko@uni-wuerzburg.de SEPTEMBER 9 13, 2019 CROSSING Summer School on Sustainable Security & Privacy

2.37k views • 187 slides
Debugging & Logging Java Logging Java has built-in support for logging Logs contain

Debugging & Logging Java Logging Java has built-in support for logging Logs contain

Debugging & Logging Java Logging Java has built-in support for logging Logs contain messages that provide information to Software developers (e.g., debugging) System administrators Customer support agents Programs send

254 views • 10 slides
TomskGAZPROMgeofjzika Company Profjle { Logging while drilling { Production logging for reservoir

TomskGAZPROMgeofjzika Company Profjle { Logging while drilling { Production logging for reservoir

TomskGAZPROMgeofjzika Geophysical Services We provide full range of well logging and mud logging op- erations, cement quality control, MWD and well engineering services throughout Western and Eastern Siberia. Over 15 years of top quality

554 views • 31 slides
Position Paper: Challenges Towards Securing Hardware-assisted Execution Environments Zhenyu Ning 1

Position Paper: Challenges Towards Securing Hardware-assisted Execution Environments Zhenyu Ning 1

Position Paper: Challenges Towards Securing Hardware-assisted Execution Environments Zhenyu Ning 1 Fengwei Zhang 1 Weisong Shi 1 Larry Shi 2 1 Wayne State University 2 University of Houston November 21, 2019 1 Overview of The Talk Motivation

575 views • 24 slides
LHC LOGGING Timeline of t he proj ect , resources Cont ext : where does logging f it in? Basic

LHC LOGGING Timeline of t he proj ect , resources Cont ext : where does logging f it in? Basic

Outline Launching of t he LHC Logging proj ect Mandat e, scope and obj ect ives LHC LOGGING Timeline of t he proj ect , resources Cont ext : where does logging f it in? Basic f unct ionalit ies Filt ering of dat a I nt erf acing t o ot

667 views • 3 slides
Hardware Design and Analysis of Efficient Loop Coarsening and Border Handling for Image

Hardware Design and Analysis of Efficient Loop Coarsening and Border Handling for Image

Hardware Design and Analysis of Efficient Loop Coarsening and Border Handling for Image Processing M. Akif zkan, Oliver Reiche, Frank Hannig, and Jrgen Teich Hardware/Software Co-Design, Friedrich-Alexander University Erlangen-Nrnberg

891 views • 65 slides
DATABASE SYSTEM IMPLEMENTATION GT 4420/6422 // SPRING 2019 // @JOY_ARULRAJ LECTURE #5: LOGGING

DATABASE SYSTEM IMPLEMENTATION GT 4420/6422 // SPRING 2019 // @JOY_ARULRAJ LECTURE #5: LOGGING

DATABASE SYSTEM IMPLEMENTATION GT 4420/6422 // SPRING 2019 // @JOY_ARULRAJ LECTURE #5: LOGGING AND RECOVERY PROTOCOLS 2 TODAYS AGENDA Logging Schemes Crash Course on ARIES protocol Physical Logging Logical Logging 3 LOGGING &

1.64k views • 125 slides
HDFI: Hardware-Assisted Data-flow Isolation Chengyu Song 1 , Hyungon Moon 2 , Monjur Alam 1 ,

HDFI: Hardware-Assisted Data-flow Isolation Chengyu Song 1 , Hyungon Moon 2 , Monjur Alam 1 ,

HDFI: Hardware-Assisted Data-flow Isolation Chengyu Song 1 , Hyungon Moon 2 , Monjur Alam 1 , Insu Yun 1 , Byoungyoung Lee 1 , Taesoo Kim 1 , Wenke Lee 1 , Yunheung Paek 2 1 Georgia Institute of Technology 2 Seoul National University Memory

311 views • 29 slides
SoK: A Study of Using Hardware- assisted Isolated Execu<on Environments for Security Fengwei

SoK: A Study of Using Hardware- assisted Isolated Execu<on Environments for Security Fengwei

SoK: A Study of Using Hardware- assisted Isolated Execu<on Environments for Security Fengwei Zhang Wayne State University Detroit, Michigan, USA Wayne State University CSC 6991 Topics in Computer Security 1 Overview Of The Talk

795 views • 28 slides
Welcome! Todays Agenda: Introduction Hardware Trust No One / An Efficient

Welcome! Todays Agenda: Introduction Hardware Trust No One / An Efficient

/INFOMOV/ Optimization & Vectorization J. Bikker - Sep-Nov 2018 - Lecture 12: Multithreading Welcome! Todays Agenda: Introduction Hardware Trust No One / An Efficient Pattern Experiments Final

517 views • 31 slides
Hardware Assisted Tracing on ARM with CoreSight and OpenCSD Mathieu Poirier In this Presentation

Hardware Assisted Tracing on ARM with CoreSight and OpenCSD Mathieu Poirier In this Presentation

Hardware Assisted Tracing on ARM with CoreSight and OpenCSD Mathieu Poirier In this Presentation End-to-end overview of the technology Not an in-depth presentation on CoreSight Emphasis on how to use rather than what it is

470 views • 35 slides
Shared Variable Proof Methods, Hardware-Assisted Critical Sections Dr. Liam OConnor University

Shared Variable Proof Methods, Hardware-Assisted Critical Sections Dr. Liam OConnor University

Transition Diagrams Owicki-Gries Method Invariants and Machine Instructions Shared Variable Proof Methods, Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW) Term 2 2020 1 Transition Diagrams

954 views • 49 slides
Shared Variable Proof Methods, Hardware-Assisted Critical Sections Dr. Liam OConnor University

Shared Variable Proof Methods, Hardware-Assisted Critical Sections Dr. Liam OConnor University

Transition Diagrams Owicki-Gries Method Invariants and Machine Instructions Shared Variable Proof Methods, Hardware-Assisted Critical Sections Dr. Liam OConnor University of Edinburgh LFCS (and UNSW) Term 2 2020 1 Transition Diagrams

162 views • 12 slides
Storage Efficient Hardware Prefetching using Delta Correlating Prediction Tables Marius Granns

Storage Efficient Hardware Prefetching using Delta Correlating Prediction Tables Marius Granns

Storage Efficient Hardware Prefetching using Delta Correlating Prediction Tables Marius Granns Magnus Jahre Lasse Natvig Feb 14th 2008 www.ntnu.no M. Granns et.al., Storage Efficient Hardware Prefetching using Delta Correlating Prediction

658 views • 52 slides

More recommend