HOW TO SCALE UP WITHOUT GIVING UP

TRANSACTIONS?

High-Performance Transaction Processing System

MAIN MEMORY STORAGE

SERIAL EXECUTION

COMMAND LOGGING

STORED PROCEDURES

0

25,000

50,000

75,000

100,000

125,000

1 2 3 4 5 6 7 8

H-Store

13x

TXN/SEC CPU CORES

MySQL Postgres

Japanese “American Idol” VOTER BENCHMARK

DISTRIBUTED TRANSACTIONS

TXN/SEC NODES (8 CORES PER NODE)

10% Distributed Transactions VOTER BENCHMARK

0

20,000

40,000

60,000

80,000

100,000

1 2 3 4

H-Store

Identify Distributed Transactions ON PREDICTIVE MODELING FOR OPTIMIZING TRANSACTION EXECUTION IN PARALLEL OLTP SYSTEMS Proc. VLDB Endow., vol. 5, pp. 85-96, 2011.

Utilize Transaction Stalls THE ART OF SPECULATIVE EXECUTION In Progress (July 2014)

Optimize Single-Partition Execution H-STORE: A HIGH-PERFORMANCE, DISTRIBUTED MAIN MEMORY TRANSACTION PROCESSING SYSTEM Proc. VLDB Endow., vol. 1, iss. 2, pp. 1496-1499, 2008.

Minimize Distributed Transactions SKEW-AWARE AUTOMATIC DATABASE PARTITIONING IN SHARED-NOTHING, PARALLEL OLTP SYSTEMS Proceedings of SIGMOD, pp. 61-72, 2012.

TXN/SEC NODES (8 CORES PER NODE)

All Optimizations Enabled VOTER BENCHMARK

0

100,000

200,000

300,000

400,000

500,000

1 2 3 4

H-Store

Stan Zdonik

Mike Stonebraker

Sam Madden

Ugur Cetintemel

David DeWitt

Dan Abadi

Justin DeBrabant

Saurya Velagapudi

Aaron Elmore

Xin Jia

Carlo Curino

Evan Jones

Yang Zou

Ning Shi

Visawee Angkana.

@ANDY_PAVLO END

db.cs.cmu.edu