STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE...
Transcript of STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE...
![Page 1: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/1.jpg)
1
![Page 2: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/2.jpg)
2
STORAGE LATENCY2
RAMAC 350(600 ms)
1956
NAND SSD(60 us)
2016
105x
![Page 3: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/3.jpg)
3
COMPUTE LATENCY3
CORE I7(1 GHZ)
2016
RAMAC 305(100 Hz)
1956
108x1000x
![Page 4: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/4.jpg)
4
NON-VOLATILE MEMORY4
1000x faster than NAND
10x denser than DRAM
1000x endurance of NAND
2017
3D XPOINT(60 ns)
![Page 5: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/5.jpg)
5
NON-VOLATILE MEMORY5
3D XPOINT(60 ns)
2017
Support in Linux 4.3+
New assembly instructions
SNIA programming model
![Page 6: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/6.jpg)
6
• Option 1: Treat NVM like a faster SSD
6
DRAM
NVM
DBMS
Improves performance
Use NVM as a logging and backing store
WHAT NVM MEANS FOR DATABASES?
![Page 7: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/7.jpg)
7
WHAT NVM MEANS FOR DATABASES?
• Option 2: Treat NVM as extended memory
7
DRAM NVM
DBMS
Improves not only performance,but also availability
and device utilization
Redesign the key algorithms in a database system
![Page 8: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/8.jpg)
88
PAST:EXISTINGSYSTEMS
PRESENT:NVM-AWARE
DBMS
FUTURE:ANALYTICS
ON NVM
PAST:EXISTINGSYSTEMS
PRESENT:NVM-AWARE
DBMS
FUTURE:ANALYTICS
ON NVM
![Page 9: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/9.jpg)
9
PAST — EXISTING SYSTEMS• Investigate how existing systems perform on NVM– Option 1: Treat NVM like a faster SSD
• Evaluate two types of DBMSs– Disk-oriented DBMS– In-memory DBMS
9
A PROLEGOMENON ON OLTP DATABASE SYSTEMS FOR NON-VOLATILE MEMORYADMS 2014
![Page 10: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/10.jpg)
10
DBMS ARCHITECTURES10
DISK-ORIENTED DBMS IN-MEMORY DBMS
Table Heap
LogCheckpoints
Buffer Pool Table Heap
LogCheckpoints
DRAM
NVM
DRAM
NVM
![Page 11: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/11.jpg)
11
NVM HARDWARE EMULATOR
• Tunable DRAM latency for emulating NVM• Special assembly instructions for NVM– Cache line write-back
11
L1 Cache
L2 Cache
STORE CLWB
CPU NVMCACHE
![Page 12: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/12.jpg)
12
EVALUATION
• Compare existing DBMSs on NVM emulator– MySQL (Disk-oriented DBMS)– H-Store (In-memory DBMS)
• TPC-C benchmark– 1/8th of database fits in DRAM, rest on NVM
12
![Page 13: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/13.jpg)
13
0
20,000
40,000
Database systems
PERFORMANCE13
In-memory DBMSDisk-Oriented DBMS
Throughput(txn/sec)
8x DRAM Latency
12x
![Page 14: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/14.jpg)
14
0
20,000
40,000
Database systems
PERFORMANCE14
Throughput(txn/sec)
In-memory DBMSDisk-Oriented DBMS
2x DRAM Latency 4x
1x
1xLegacy database systems are not prepared for NVM
![Page 15: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/15.jpg)
1515
PAST:EXISTINGSYSTEMS
PRESENT:NVM-AWARE
DBMS
FUTURE:ANALYTICS
ON NVM
PRESENT:NVM-AWARE
DBMS
![Page 16: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/16.jpg)
16
PRESENT — NVM-AWARE DBMS16
• Understand changes required in DBMSs to leverage NVM– Option 2: Treat NVM as extended memory
• Rethink key algorithms in database systems– Logging and recovery protocol
• Designed for real-time analytics– Multi-versioned database
LET’S TALK ABOUT STORAGE AND RECOVERY METHODS FORNON-VOLATILE MEMORY DATABASE SYSTEMSSIGMOD 2015
![Page 17: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/17.jpg)
17
MULTI-VERSIONED DBMS17
BEGINTIMESTAMP
ENDTIMESTAMP
PREVIOUSVERSION
TUPLE ID
TUPLEDATA
10 ∞ —1 X
10 ∞ —2 Y
20 ∞ 23 Y’
10 20 —
![Page 18: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/18.jpg)
18
THOUGHT EXPERIMENT
• NVM-only storage hierarchy– No volatile DRAM
18
NVM
DBMS
![Page 19: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/19.jpg)
19
TRADITIONAL STORAGE ENGINE19
Table Heap
LogCheckpoints
1
2 3NVM
DataData
![Page 20: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/20.jpg)
20
NON-VOLATILE POINTER20
POINTER DATA
DRAM DRAM
POINTER DATA
NVM NVM
VOLATILE POINTER NON-VOLATILE POINTER
CRASH: DISAPPEARS CRASH: VALID
ü✗
![Page 21: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/21.jpg)
21
NVM-AWARE STORAGE ENGINE
• Instead of duplicating tuple data in log– Store non-volatile tuple pointers in log records
21
TRADITIONAL LOG
INSERT TUPLE 1 (DATA)
UPDATE TUPLE 1 (DATA)
NVM-AWARE LOG
INSERT TUPLE 1 (POINTER)
UPDATE TUPLE 1 (POINTER)
![Page 22: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/22.jpg)
22
NVM-AWARE STORAGE ENGINE22
Table Heap
Log
1
2NVM
MetaData
Checkpoints✗
![Page 23: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/23.jpg)
23
EVALUATION
• Compare storage engines on NVM emulator– Traditional engine– NVM-aware engine
• Yahoo! Cloud Serving Benchmark– Database fits in NVM
23
![Page 24: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/24.jpg)
24
PERFORMANCE24
Throughput(txn/sec)
0
300,000
600,000
Storage Engines
3x
NVM-Aware EngineTraditional Engine
8x DRAM Latency
![Page 25: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/25.jpg)
2525
Throughput(txn/sec)
0
750,000
1,500,000
Storage Engines
6x
PERFORMANCENVM-Aware EngineTraditional Engine
2x DRAM Latency 4x
1x
NVM latency has a significant impact on the performance of NVM-aware storage engine
![Page 26: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/26.jpg)
26
DEVICE LIFETIME26
NVM Stores(M)
0
50
100
Storage Engines
2x
NVM-Aware EngineTraditional Engine
![Page 27: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/27.jpg)
27
THE PELOTON DBMS
• A new database system for NVM research – Designed for a two-tier storage hierarchy
27
DRAM NVM
DBMS
WRITE-BEHIND LOGGINGUNDER REVIEW
![Page 28: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/28.jpg)
28
WRITE-BEHIND LOGGING
• Write-ahead log serves two purposes– Transform random database writes into sequential log writes– Support transaction rollback
• NVM supports fast random writes– Directly write data to the database– Later, record metadata in write-behind log
28
![Page 29: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/29.jpg)
29
WRITE-BEHIND LOGGING29
Database2
Database
Log
1
3
DRAM
NVM
MetaData Data
![Page 30: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/30.jpg)
30
METADATA FOR INSTANT RECOVERY
• Record failed timestamp ranges in log– Use it to ignore effects of uncommitted transactions
30
Group Commit
Time
(T1, T2) (T1, T2)
Garbage Collection
Dirty Ranges
T1 T4T3T2
![Page 31: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/31.jpg)
31
EVALUATION
• Compare logging protocols in Peloton– Write-Ahead logging– Write-Behind logging
• Storage devices in Intel’s NVM hardware emulator– Hard-disk drive– Solid-state drive– Non-volatile memory emulator
• TPC-C benchmark
31
![Page 32: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/32.jpg)
32
APPLICATION AVAILABILITY32
1
100
10,000
Hard Disk Drive
Solid State Drive
Non-Volatile Memory
Write-Behind LoggingWrite-Ahead Logging
Recovery Time(sec)
1000x
![Page 33: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/33.jpg)
33
PERFORMANCE33
Write-Behind LoggingWrite-Ahead Logging
Throughput(txn/sec)
1
100
10,000
Hard Disk Drive
Solid State Drive
Non-Volatile Memory
10x
1.3x
![Page 34: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/34.jpg)
3434
PAST:EXISTINGSYSTEMS
PRESENT:NVM-AWARE
DBMS
FUTURE:ANALYTICS
ON NVM
FUTURE:ANALYTICS
ON NVM
![Page 35: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/35.jpg)
35
FUTURE — REAL-TIME ANALYTICS
• Support analytics on larger-than-NVM databases– Cost of first-generation NVM devices
• Three-tier storage hierarchy– DRAM + NVM + SSD
• When to migrate data between different layers?
35
REAL-TIME ANALYTICS IN NON-VOLATILE MEMORY DATABASE SYSTEMS WORK IN PROGRESS
![Page 36: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/36.jpg)
36
THREE-TIER STORAGE HIERARCHY36
Database2
Database
Log
1
3
DRAM
NVM
Database4
SSD
Data
Data
MetaData
![Page 37: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/37.jpg)
37
DATA MIGRATION
• Can directly read warm data from NVM– No need to copy data over to DRAM for reading
• Keep hottest data in DRAM• Dynamically migrate cold data to SSD
37
![Page 38: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/38.jpg)
38
THREE-TIER STORAGE HIERARCHY38
0
60,000
120,000
Read-Heavy Workload
Balanced Workload
Write-Heavy Workload
Throughput(txn/sec)
50% on NVM10% on NVM 90% on NVM
8x3x 4x
![Page 39: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/39.jpg)
3939
PAST:EXISTINGSYSTEMS
PRESENT:NVM-ORIENTED
DBMS
FUTURE:ANALYTICS
ON NVM
PAST:EXISTINGSYSTEMS
PRESENT:NVM-AWARE
DBMS
FUTURE:ANALYTICS
ON NVM
![Page 40: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/40.jpg)
40
LESSONS LEARNED
• NVM outperforms SSD when correctly used • Rethink key algorithms in database systems– Write-behind logging enables instant recovery– Improves device utilization and extends its lifetime
• Ongoing work– Data placement policy– Replication
40
![Page 41: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/41.jpg)
4141
![Page 42: STORAGE LATENCYjarulraj/talks/2016.wbl.pdf · 2018-09-25 · Storage Engines 6x PERFORMANCE Traditional Engine NVM-Aware Engine 2x DRAM Latency 4x 1x NVM latency has a significant](https://reader034.fdocuments.net/reader034/viewer/2022050507/5f98daff259ee834581d034f/html5/thumbnails/42.jpg)
4242