Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686...
Transcript of Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686...
![Page 1: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/1.jpg)
Systems | Fueling future disruptions
ResearchFaculty Summit 2018
![Page 2: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/2.jpg)
@andy_pavlo[Image Source]
![Page 3: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/3.jpg)
Part #1Part #2Part #3
– Background– Engineering– Oracle Rant
![Page 4: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/4.jpg)
AUTONOMOUS DBMSs 4SELF-ADAPTIVE DATABASES
1970-1990sSelf-Adaptive
Databases
SELECT * FROM A JOIN BON A.ID = B.ID
WHERE A.VAL > 123AND B.NAME LIKE 'XY%'
A.IDA.VALB.ID
B.NAME
+100 +200 +50
TuningAlgorithm
Admin
→Index Selection→Partitioning / Sharding→Data Placement
![Page 5: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/5.jpg)
541
291
0
200
400
600
2000 2004 2008 2012 2016
Number of Knobs
AUTONOMOUS DBMSs 5SELF-TUNING DATABASES
1990-2000sSelf-TuningDatabases
SELECT * FROM A JOIN BON A.ID = B.ID
WHERE A.VAL > 123AND B.NAME LIKE 'XY%'
TuningAlgorithm
Admin
→Index Selection→Partitioning / Sharding→Data PlacementAutoAdmin
A.IDA.VALB.ID
B.NAME
→Knob Configuration
![Page 6: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/6.jpg)
AUTONOMOUS DBMSs 6CLOUD MANAGED DATABASES
2010sCloud
Databases
→Initial Placement→Tenant Migration
![Page 7: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/7.jpg)
Why is this Previous WorkInsufficient?
![Page 8: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/8.jpg)
AUTONOMOUS DBMSs 8A BRIEF HISTORY
Problem #1Human
Judgements
Problem #2ReactionaryMeasures
Problem #3No Transfer
Learning
![Page 9: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/9.jpg)
What is Different this Time?
![Page 10: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/10.jpg)
CARNEGIE MELLON UNIVERSITY 10RESEARCH PROJECTS
OtterTuneExistingSystems
PelotonNew
System
![Page 11: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/11.jpg)
OtterTune
Database Tuning-as-a-Service→Automatically generate
DBMS knob configurations.→Reuse data from previous
tuning sessions.
ottertune.cs.cmu.edu
![Page 12: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/12.jpg)
OTTERTUNETPC-C TUNING
AUTOMATIC DATABASE MANAGEMENT SYSTEM TUNING THROUGH LARGE-SCALE MACHINE LEARNINGSIGMOD 2017
165
508562
736686
0
250
500
750
1000
426
845
714
843
946
0
250
500
750
1000Throughput (txn/sec)
Default RDS DBAScripts OtterTune
12
![Page 13: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/13.jpg)
Peloton
Self-Driving Database System→ In-memory DBMS with
integrated planning framework.
→Designed for autonomous operations.
pelotondb.io
![Page 14: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/14.jpg)
Design Considerations forAutonomous Operation
![Page 15: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/15.jpg)
AUTONMOUS DBMS 15DESIGN CONSIDERATIONS
ConfigurationKnobs
InternalMetrics
ActionEngineering
![Page 16: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/16.jpg)
Anything that requires a human value judgement should be marked as off-limits to autonomous components.
– File Paths / Network Info– Durability / Isolation Levels– Hardware Usage– Recovery Time
16UNTUNABLE KNOBSCONFIGURATION KNOBS
![Page 17: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/17.jpg)
CONFIGURATION KNOBS
The autonomous components need hints about how to change a knob.
– Min/max ranges.– Separate knobs to enable/disable a feature.– Non-uniform deltas.
17HOW TO CHANGE
1 KB 1 MB 1 GB 1 TB+100 KB +100 MB +100 GB
![Page 18: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/18.jpg)
INTERNAL METRICS
If the DBMS has sub-components that are tunable, then it must expose separate metrics for those components.
Bad Example:
18SUB-COMPONENTS
![Page 19: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/19.jpg)
INTERNAL METRICS 19SUB-COMPONENTS
RocksDB Column Family Knobs
Column Family Metrics
Missing:ReadsWrites
Global Metrics
AggregatedMetrics
![Page 20: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/20.jpg)
ACTION ENGINEERING
No action should ever require the DBMS to restart in order for it to take affect.The commercial systems are much better than this than the open-source systems.
20NO SHUTDOWN
![Page 21: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/21.jpg)
ACTION ENGINEERING
Allow replica configurations to diverge from each other.
21REPLICA EXPLORATION
Master
Replicas
![Page 22: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/22.jpg)
What About Oracle'sSelf-Driving DBMS?
![Page 23: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/23.jpg)
ORACLE
Automatic Indexing
Automatic Recovery
Automatic Scaling
Automatic Query Tuning
SELF-DRIVING DBMS23
September 2017 January 2017
Problem #2React ionary
Measures
![Page 24: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/24.jpg)
CONCLUSION
True autonomous DBMSs are achievable in the next decade.You should think about how each new feature can be controlled by a machine.
MAIN TAKEAWAYS24
![Page 26: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/26.jpg)
![Page 27: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/27.jpg)
OTTERTUNE 27AUTOMATIC DBMS TUNING SERVICE
TARGETDATABASE
TUNING MANAGERCONTROLLERCOLLECTOR
Configuration Recommender
KnobAnalyzer
MetricAnalyzer
5001000150020002500
200400
600800
99th
%-ti
le(s
ec)
0.00.51.01.52.0
InternalRepository
INSTALL AGENT
![Page 28: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/28.jpg)
"THE BRAIN"PELOTON 28THE SELF-DRIVING DBMS
FORECAST MODELS
Search Tree
ACTIONCATALOG
ACTION SEQUENCETARGET
DATABASE
WORKLOADHISTORY ?
??
QUERY-BASED WORKLOAD FORECASTING FOR SELF-DRIVING DATABASE MANAGEMENT SYSTEMSIGMOD 2018
![Page 29: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/29.jpg)
PELOTONBUS TRACKING APP WITH ONE-HOUR HORIZON
QUERY-BASED WORKLOAD FORECASTING FOR SELF-DRIVING DATABASE MANAGEMENT SYSTEMSIGMOD 2018
0
15000
30000
45000
60000
9-Jan 11-Jan 13-Jan 15-Jan 17-Jan
Ensemble (LR+RNN)
29
Actual Predicted
Que
ries
Per H
our
![Page 30: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/30.jpg)
ACTION ENGINEERING
Provide a notification callback to indicate when an action starts and when it completes.Harder for changes that can be used before the action completes.
30NOTIFICATIONS
![Page 31: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/31.jpg)
Thank you!
![Page 32: Research Faculty Summit 2018...THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017 165 508 562 736 686 0 250 500 750 1000 426 845 714 843 946 0 250 500 750 1000 Throughput (txn/sec) Default](https://reader035.fdocuments.us/reader035/viewer/2022070723/5f01fda97e708231d402093a/html5/thumbnails/32.jpg)