Pebbles DB: Building Key-Value Stores using Fragmented Log- - - PowerPoint PPT Presentation

Pebbles DB: Building Key-Value Stores using Fragmented Log- Structured Merge Trees(II)

Peter Sassaman

Overview

• Sentinel
• Guards
• Guard Selection (MurmurHash)
• SSTables
• High Levels in memory

(1) “FLSM can be viewed as a generalization of the LSM data structure.”. Please explain this statement.

(2)“New sstables are simply added to the correct guard in the next level. There are two exceptions to the no-rewrite rule”. Describe these two exception compaction scenarios.

Level 4 Level 5

1,2 1,2

Level 3 Level 4 Level 5

1,2 1,2 1,2 An SSTAble An SSTAble Guard on this level almost full

(3)“deleting a guard will involve a significant amount of compaction work” Please describe the

• perations involved in a guard deletion.
• At Compaction
• add to in-memory set
• SS Tables are Either appended to neighbor guard in same level or

child guards in next level.

• Keep metadata
• If needed delete at levels > i
• delete levels < i

(4) Describe how a range query is served.

Query: Get 2- 5

(4) Describe how a range query is served.

Query Get 2 - 5

• 1. Guard: 1,5

(4) Describe how a range query is served.

Get 2 - 5

• 1. Guard: 1,5
• 2. 2. Binary

2

(4) Describe how a range query is served.

Get 2 - 5

• 1. Guard: 1,5
• 2. Binary
• 3. 3. Merge

2,3

(4) Describe how a range query is served.

Get 2 - 5

• 1. Guard: 1,5
• 2. Binary
• 3. Merge
• 4. End

2,3,5

End of Query

(5) “In Workload F, all writes are read-modify-writes: the workload does a get() before every put() operation. As a result, the full write throughput of PebblesDb is not utilized, resulting in performance similar to that of other key-value stores.” Explain why for this workload PebbleDB has similar performance.

Images From

• http://www.cs.utexas.edu/~vijay/papers/sosp17-pebblesdb.pdf
• https://upload.wikimedia.org/wikipedia/commons/thumb/f/f2/LSM_

Tree.png/1200px-LSM_Tree.png