1. SILT: A Memory-Efficient, High-Performance Key-Value Store
Hyeontaek Lim, Bin Fan, Davig G. Anderson, Michael Kaminsky*
CMU, *Intel Lab
Presenter: Taejoong Chung
*Some slides are borrowed from author’s one

2. Contents Introduction Objective Wish to think in this seminar
Struggles in Key-Value Storage
Objective
Design Goal
Wish to think in this seminar
Prerequisites
Methodology
LogStore, HashStore, SortedStore
Conclusion

3. Introduction: Struggles in Key-Value Storages
Because of economical/speed issues, storages for key and values are separated
Basically, keys are mainly stored in fast-memory (e.g., S/DRAM) and values are stored in slow memory (e.g., Flash)
However, flash-memory has threshold for writing
 The number of writings are carefully considered

4. Seesaw Games ? FAWN-DS How can we improve? FlashStore HashCache
SkimpyStash
BufferHash
Memory efficiency
High performance

5. Objectives The size for key in DRAM are too big, so far!
Some of keys are also stored in flash
Many flash reads for finding index bring a die young
Let’s minimize
Size for key in fast memory
Number of look-up in slow memory

6. Design Goals Low read amplification
Minimize the flash-reads (1 + ε)
Controllable write amplification and favoring sequential writes
Also minimize the flash-writes  mortality
Memory-efficient indexing
Computation-efficient indexing
Effective use of flash spaces

7. Wish to Think in This Seminar
How to apply this structures/mechanism for NDN components CS
PIT
FIB
But, because flash memory are too slow to get the data back
Fast + Slow memory  CS (Content Store)
Fast memory  FIB or PIT

8. Prerequisite: Cuckoo Hashing
Name derives from the behavior of cuckoo-chick who pushes the other eggs when it hatches
Probing problems in basic-hashing
If there is lots of entries with same hash function, the bucket size increases, number of probing is increased
Two hash functions are used for an index
If another elements are inserted, the original elements are shifted to the other index

9. Prerequisite: Compressed Trie
The representation for tries have many alternatives
Compressed Trie

10. Approaches (Conventional vs. SILT)
Conventional Structure
In-memory filter to test whether a given key is stored before accessing flashes
In-memory index to locate the data on flash
On-flash data to store all key-values
SILT Structure (3 components)

11. Basic Operations LogStore HashStore SortedStore
Write friendly key-value store (similar to conventional storage)
In-memory hash table to map contents in flash
Modified Cuckoo hashing is used
Only one instance
HashStore
Immutable hash store
No need to in-memory index to locate entries (what?!)
SortedStore
Key-value storage on flash
Extremely compact index representation (0.4 bytes/key)

12. #1. LogStore
Use a tag (15 bits) for an index rather than an entire hash index (160bits)
Cuckoo hashing is used
When it happens to replacement?
Should find alternative-hash for an originally inserted hash  another read from flash
Insert the other hash result for an tag

13. #2. HashStore
Once a LogStore fills up, SILT freezes the LogStore and converts it into HashStore
Sort the LogStore based on ‘HashOrder’
Saves lots of in-memory by eliminating the index and reordering the on-flash pairs from insertion order to hash-order

14. Where are We?

15. #3. SortedStore Multiple HashStore can be aggregated into SortedStore
Struggle to minimize the bit presentation
From the sorted results, indices are re-made

16. SLIT Lookup Process
Queries look up stores in sequence (from new to old)
Inserts only go to Log
Data are moved in background
SILT Sorted Index
(Memory efficient)
SILT Filter
SILT Log Index
(Write friendly)
Memory
Flash

17. SILT’s Design <SortedStore> <HashStore> <LogStore>
SILT Sorted Index
SILT Filter
SILT Log Index
Merge
Conversion
On-flash sorted array
On-flash hashtables
On-flash log
Memory overhead
Read amplification
Write amplification
0.7 bytes/entry
1.01
5.4

18. Conclusion
SILT provides both memory-efficient and high-performance key-value store
Multi-store approach
Entropy-coded tries
Partial-key cuckoo hashing
Full source code is available