Ad
More Related Content
What's hot (20)
Viewers also liked (15)
Ad
Similar to Native erasure coding support inside hdfs presentation (20)
Ad
Recently uploaded (20)
Native erasure coding support inside hdfs presentation
- 1. HDFS Erasure Coding Zhe Zhang [email protected]
- 3. § HDFS inherits 3-way replication from Google File System - Simple, scalable and robust Replication is Expensive Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica
- 4. § HDFS inherits 3-way replication from Google File System - Simple, scalable and robust § 200% storage overhead Replication is Expensive Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica
- 5. § HDFS inherits 3-way replication from Google File System - Simple, scalable and robust § 200% storage overhead § Secondary replicas rarely accessed Replication is Expensive Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica
- 6. Erasure Coding Saves Storage
- 7. Erasure Coding Saves Storage § Simplified Example: storing 2 bits 1 0Replication: XOR Coding: 1 0
- 8. Erasure Coding Saves Storage § Simplified Example: storing 2 bits 1 01 0Replication: XOR Coding: 1 0
- 9. Erasure Coding Saves Storage § Simplified Example: storing 2 bits 1 01 0Replication: XOR Coding: 1 0 2 extra bits
- 10. Erasure Coding Saves Storage § Simplified Example: storing 2 bits 1 01 0Replication: XOR Coding: 1 0⊕ 1= 2 extra bits
- 11. Erasure Coding Saves Storage § Simplified Example: storing 2 bits 1 01 0Replication: XOR Coding: 1 0⊕ 1= 2 extra bits 1 extra bit
- 12. Erasure Coding Saves Storage § Simplified Example: storing 2 bits § Same data durability - can lose any 1 bit 1 01 0Replication: XOR Coding: 1 0⊕ 1= 2 extra bits 1 extra bit
- 13. Erasure Coding Saves Storage § Simplified Example: storing 2 bits § Same data durability - can lose any 1 bit § Half the storage overhead 1 01 0Replication: XOR Coding: 1 0⊕ 1= 2 extra bits 1 extra bit
- 14. Erasure Coding Saves Storage § Simplified Example: storing 2 bits § Same data durability - can lose any 1 bit § Half the storage overhead § Slower recovery 1 01 0Replication: XOR Coding: 1 0⊕ 1= 2 extra bits 1 extra bit
- 15. Erasure Coding Saves Storage
- 16. Erasure Coding Saves Storage § Facebook - f4 stores 65PB of BLOBs in EC
- 17. Erasure Coding Saves Storage § Facebook - f4 stores 65PB of BLOBs in EC § Windows Azure Storage (WAS) - A PB of new data every 1~2 days - All “sealed” data stored in EC
- 18. Erasure Coding Saves Storage § Facebook - f4 stores 65PB of BLOBs in EC § Windows Azure Storage (WAS) - A PB of new data every 1~2 days - All “sealed” data stored in EC § Google File System - Large portion of data stored in EC
- 19. Roadmap
- 20. Roadmap § Background of EC - Redundancy Theory - EC in Distributed Storage Systems
- 21. Roadmap § Background of EC - Redundancy Theory - EC in Distributed Storage Systems § HDFS-EC architecture - Choosing Block Layout - NameNode — Generalizing the Block Concept - Client — Parallel I/O - DataNode — Background Reconstruction
- 22. Roadmap § Background of EC - Redundancy Theory - EC in Distributed Storage Systems § HDFS-EC architecture - Choosing Block Layout - NameNode — Generalizing the Block Concept - Client — Parallel I/O - DataNode — Background Reconstruction § Hardware-accelerated Codec Framework
- 23. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data?
- 24. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica 3-way Replication:
- 25. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica 3-way Replication:
- 26. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica 3-way Replication: Data Durability = 2
- 27. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica 3-way Replication: Data Durability = 2
- 28. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica useful data 3-way Replication: Data Durability = 2 redundant data
- 29. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Replica DataNode0 DataNode1 DataNode2 Block NameNode Replica Replica useful data 3-way Replication: Data Durability = 2 Storage Efficiency = 1/3 (33%) redundant data
- 30. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data?
- 31. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? XOR: X Y X ⊕ Y 0 0 0 0 1 1 1 0 1 1 1 0
- 32. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? XOR: X Y X ⊕ Y 0 0 0 0 1 1 1 0 1 1 1 0 Y = 0 ⊕ 1 = 1
- 33. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? XOR: Data Durability = 1 X Y X ⊕ Y 0 0 0 0 1 1 1 0 1 1 1 0 Y = 0 ⊕ 1 = 1
- 34. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? XOR: Data Durability = 1 useful data redundant data X Y X ⊕ Y 0 0 0 0 1 1 1 0 1 1 1 0
- 35. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? XOR: Data Durability = 1 Storage Efficiency = 2/3 (67%) useful data redundant data X Y X ⊕ Y 0 0 0 0 1 1 1 0 1 1 1 0
- 36. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Reed-Solomon (RS):
- 37. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Reed-Solomon (RS):
- 38. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Reed-Solomon (RS): Data Durability = 2 Storage Efficiency = 4/6 (67%)
- 39. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Reed-Solomon (RS): Data Durability = 2 Storage Efficiency = 4/6 (67%) Very flexible!
- 40. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data?
- 41. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency
- 42. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica
- 43. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0
- 44. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100%
- 45. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication
- 46. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2
- 47. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33%
- 48. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells
- 49. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1
- 50. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1 86%
- 51. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1 86% RS (6,3)
- 52. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1 86% RS (6,3) 3
- 53. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1 86% RS (6,3) 3 67%
- 54. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1 86% RS (6,3) 3 67% RS (10,4)
- 55. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1 86% RS (6,3) 3 67% RS (10,4) 4
- 56. Durability and Efficiency Data Durability = How many simultaneous failures can be tolerated? Storage Efficiency = How much portion of storage is for useful data? Data Durability Storage Efficiency Single Replica 0 100% 3-way Replication 2 33% XOR with 6 data cells 1 86% RS (6,3) 3 67% RS (10,4) 4 71%
- 57. EC in Distributed Storage Block Layout: 128~256MFile 0~128M … 640~768M0~128M 128~256M
- 58. EC in Distributed Storage Block Layout: 128~256MFile … 640~768M 0~128 M block0 DataNode 0 0~128M 128~256M
- 59. EC in Distributed Storage Block Layout: File … 640~768M 0~128 M block0 DataNode 0 128~ 256M block1 DataNode 1 0~128M 128~256M
- 60. EC in Distributed Storage Block Layout: File … 640~768M 0~128 M block0 DataNode 0 128~ 256M block1 DataNode 1 0~128M 128~256M … 640~ 768M block5 DataNode 5
- 61. EC in Distributed Storage Block Layout: File … 640~768M 0~128 M block0 DataNode 0 128~ 256M block1 DataNode 1 0~128M 128~256M … 640~ 768M block5 DataNode 5 DataNode 6 … parity
- 62. EC in Distributed Storage Block Layout: File … 640~768M 0~128 M block0 DataNode 0 128~ 256M block1 DataNode 1 0~128M 128~256M … 640~ 768M block5 DataNode 5 DataNode 6 … parity Contiguous Layout:
- 63. EC in Distributed Storage Block Layout: Data Locality ! File … 640~768M 0~128 M block0 DataNode 0 128~ 256M block1 DataNode 1 0~128M 128~256M … 640~ 768M block5 DataNode 5 DataNode 6 … parity Contiguous Layout:
- 64. EC in Distributed Storage Block Layout: Data Locality ! Small Files " File … 640~768M 0~128 M block0 DataNode 0 128~ 256M block1 DataNode 1 0~128M 128~256M … 640~ 768M block5 DataNode 5 DataNode 6 … parity Contiguous Layout:
- 65. EC in Distributed Storage Block Layout: File block0 DataNode 0 block1 DataNode 1 … block5 DataNode 5 DataNode 6 … parity 0~128M 128~256M
- 66. EC in Distributed Storage Block Layout: File block0 DataNode 0 block1 DataNode 1 … block5 DataNode 5 DataNode 6 … parity 0~1M 1~2M 5~6M 0~128M 128~256M
- 67. EC in Distributed Storage Block Layout: File block0 DataNode 0 block1 DataNode 1 … block5 DataNode 5 DataNode 6 … parity 0~1M 1~2M 5~6M 6~7M 0~128M 128~256M
- 68. EC in Distributed Storage Block Layout: File block0 DataNode 0 block1 DataNode 1 … block5 DataNode 5 DataNode 6 … parity Striped Layout: 0~1M 1~2M 5~6M 6~7M Data Locality " Small Files ! Parallel I/O ! 0~128M 128~256M
- 69. EC in Distributed Storage Spectrum: Replication Erasure Coding Striping Contiguous Ceph Ceph Quancast File System Quancast File System HDFS Facebook f4 Windows Azure
- 70. Roadmap § Background of EC - Redundancy Theory - EC in Distributed Storage Systems § HDFS-EC architecture - Choosing Block Layout - NameNode — Generalizing the Block Concept - Client — Parallel I/O - DataNode — Background Reconstruction § Hardware-accelerated Codec Framework
- 71. Choosing Block Layout • Medium: 1~6 blocks• Small files: < 1 block• Assuming (6,3) coding • Large: > 6 blocks (1 group)
- 72. Choosing Block Layout • Medium: 1~6 blocks• Small files: < 1 block• Assuming (6,3) coding • Large: > 6 blocks (1 group) 64.61% 9.33% 26.06% 1.85%1.86% 96.29% small medium large file count space usage Top 2% files occupy ~65% space Cluster A Profile
- 73. Choosing Block Layout • Medium: 1~6 blocks• Small files: < 1 block• Assuming (6,3) coding • Large: > 6 blocks (1 group) 64.61% 9.33% 26.06% 1.85%1.86% 96.29% small medium large file count space usage Top 2% files occupy ~65% space Cluster A Profile 40.08% 36.03% 23.89% 2.03% 11.38% 86.59% file count space usage Top 2% files occupy ~40% space small medium large Cluster B Profile
- 74. Choosing Block Layout • Medium: 1~6 blocks• Small files: < 1 block• Assuming (6,3) coding • Large: > 6 blocks (1 group) 64.61% 9.33% 26.06% 1.85%1.86% 96.29% small medium large file count space usage Top 2% files occupy ~65% space Cluster A Profile 40.08% 36.03% 23.89% 2.03% 11.38% 86.59% file count space usage Top 2% files occupy ~40% space small medium large Cluster B Profile 3.20% 20.75% 76.05% 0.00%0.36% 99.64% file count space usage Dominated by small files small medium large Cluster C Profile
- 75. Choosing Block Layout Striping Contiguous Replication Erasure Coding Phase 1.1 Phase 1.2 Phase 2 (Future work) Phase 3 (Future work) Current HDFS
- 77. Generalizing Block NameNode Mapping Logical and Storage Blocks
- 78. Generalizing Block NameNode Mapping Logical and Storage Blocks Too Many Storage Blocks?
- 79. Generalizing Block NameNode Mapping Logical and Storage Blocks Too Many Storage Blocks? Hierarchical Naming Protocol:
- 80. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode
- 81. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode
- 82. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode
- 83. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode
- 84. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode Coordinator
- 85. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode Coordinator
- 86. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode Coordinator
- 87. Client Parallel Writing streamer queue streamer … streamer DataNode DataNode DataNode Coordinator
- 88. Client Parallel Reading … DataNodeDataNode DataNode DataNode DataNode
- 89. Client Parallel Reading … DataNodeDataNode DataNode DataNode DataNode
- 90. Client Parallel Reading … DataNodeDataNode DataNode DataNode DataNode
- 91. Client Parallel Reading … DataNodeDataNode DataNode DataNode DataNode parity
- 92. Reconstruction on DataNode § Important to avoid delay on the critical path - Especially if original data is lost § Integrated with Replication Monitor - Under-protected EC blocks scheduled together with under-replicated blocks - New priority algorithms § New ErasureCodingWorker component on DataNode
- 93. Roadmap § Background of EC - Redundancy Theory - EC in Distributed Storage Systems § HDFS-EC architecture - Choosing Block Layout - NameNode — Generalizing the Block Concept - Client — Parallel I/O - DataNode — Background Reconstruction § Hardware-accelerated Codec Framework
- 94. Acceleration with Intel ISA-L § 1 legacy coder - From Facebook’s HDFS-RAID project § 2 new coders - Pure Java — code improvement over HDFS-RAID - Native coder with Intel’s Intelligent Storage Acceleration Library (ISA-L)
- 97. Conclusion
- 98. Conclusion § Erasure coding expands effective storage space by ~50%!
- 99. Conclusion § Erasure coding expands effective storage space by ~50%! § HDFS-EC phase I implements erasure coding in striped block layout
- 100. Conclusion § Erasure coding expands effective storage space by ~50%! § HDFS-EC phase I implements erasure coding in striped block layout § Upstream effort (HDFS-7285): - Design finalized Nov. 2014 - Development started Jan. 2015 - 218 commits, ~25k LoC change - Broad collaboration: Cloudera, Intel, Hortonworks, Huawei, Yahoo (Japan)
- 101. Conclusion § Erasure coding expands effective storage space by ~50%! § HDFS-EC phase I implements erasure coding in striped block layout § Upstream effort (HDFS-7285): - Design finalized Nov. 2014 - Development started Jan. 2015 - 218 commits, ~25k LoC change - Broad collaboration: Cloudera, Intel, Hortonworks, Huawei, Yahoo (Japan) § Phase II will support contiguous block layout for better locality
- 102. Acknowledgements § Cloudera - Andrew Wang, Aaron T. Myers, Colin McCabe, Todd Lipcon, Silvius Rus § Intel - Kai Zheng, Uma Maheswara Rao G, Vinayakumar B, Yi Liu, Weihua Jiang § Hortonworks - Jing Zhao, Tsz Wo Nicholas Sze § Huawei - Walter Su, Rakesh R, Xinwei Qin § Yahoo (Japan) - Gao Rui, Kai Sasaki, Takuya Fukudome, Hui Zheng
- 103. Just merged to trunk!
- 104. Questions? Just merged to trunk!
- 105. Questions? Just merged to trunk! Erasure Coding:A type of Error Correction Coding
- 106. EC in Distributed Storage Spectrum:
- 107. EC in Distributed Storage 0~128 M 128~256 M DataNode0 block0 block1 … DataNode1 640~768 M DataNode5 block5 Contiguous DataNode6 DataNode8 data parity … Block Layout: 128~256MFile 0~128M … 640~768M
- 108. EC in Distributed Storage 0~128 M 128~256 M DataNode0 block0 block1 … DataNode1 640~768 M DataNode5 block5 Contiguous DataNode6 DataNode8 data parity … Block Layout: Data Locality ! 128~256MFile 0~128M … 640~768M
- 109. EC in Distributed Storage 0~128 M 128~256 M DataNode0 block0 block1 … DataNode1 640~768 M DataNode5 block5 Contiguous DataNode6 DataNode8 data parity … Block Layout: Data Locality ! Small Files " 128~256MFile 0~128M … 640~768M
- 110. EC in Distributed Storage 0~128 M 128~256 M DataNode0 block0 block1 … DataNode1 640~768 M DataNode5 block5 Contiguous DataNode6 DataNode8 data parity … Block Layout: Data Locality ! Small Files " 128~256MFile … 640~768M
- 111. EC in Distributed Storage 0~1M … … 1~2M … … DataNode0 block0 DataNode1 5~6M … 127~128M DataNode5 Striping DataNode6 DataNode8 data parity …… Block Layout:
- 112. EC in Distributed Storage 0~1M … … 1~2M … … DataNode0 block0 DataNode1 5~6M … 127~128M DataNode5 Striping DataNode6 DataNode8 data parity …… Block Layout: Data Locality "
- 113. EC in Distributed Storage 0~1M … … 1~2M … … DataNode0 block0 DataNode1 5~6M … 127~128M DataNode5 Striping DataNode6 DataNode8 data parity …… Block Layout: Data Locality " Small Files !
- 114. EC in Distributed Storage 0~1M … … 1~2M … … DataNode0 block0 DataNode1 5~6M … 127~128M DataNode5 Striping DataNode6 DataNode8 data parity …… Block Layout: Data Locality " Small Files ! Parallel I/O !
- 115. Client Parallel Writing blockGroup DataStreamer 0 DataStreamer 1 DataStreamer 2 DataStreamer 3 DataStreamer 4 DFSStripedOutputStream dataQueue 0 dataQueue 1 dataQueue 2 dataQueue 3 dataQueue 4 blk_1009 blk_1010 blk_1011 blk_1012 blk_1013 Coordinator allocate new blockGroup
- 116. Client Parallel Reading Stripe 0 Stripe 1 Stripe 2 DataNode 0 DataNode 1 DataNode 2 DataNode 2 DataNode 3 (parity blocks)(data blocks) all zero all zero requested requested requested requested requested recovery read recovery read recovery read recovery read recovery read recovery read recovery read recovery read