Presentation 2006-07-14
A Mechanism for Highly-Available OLTP Systems based on PAXOS Consensus
Hiroshi HORII, Hideki TAI, Gaku YAMAMOTO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Demand for highly available (HA) on-line transaction processing system has been increasing these years. However, it is relatively difficult to make modern multi-layered systems HA because the HA mechanisms of each layer need to be coordinated carefully. In addition, service outage is dependent on failure detector which requires some amount of time until it detects a failure. We propose a mechanism that provides end-to-end HA for modern On-Line-Transaction-Processing (OLTP) systems. In our mechanism, when the primary system did not respond in a specific time, client reroutes a request to a secondary system without waiting for failure detection. Primary database server replicates transaction logs to secondary database server using PAXOS consensus algorithm. Primary and secondary database servers avoid duplicate transaction commits while they allow duplicate executions of a transaction by multiple application servers. We applied this mechanism to a Web application and evaluated it to show that it can be used for general Web applications.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Distributed Consensus / PAXOS / Replication
Paper # DE2006-111
Date of Issue

Conference Information
Committee DE
Conference Date 2006/7/7(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Data Engineering (DE)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A Mechanism for Highly-Available OLTP Systems based on PAXOS Consensus
Sub Title (in English)
Keyword(1) Distributed Consensus
Keyword(2) PAXOS
Keyword(3) Replication
1st Author's Name Hiroshi HORII
1st Author's Affiliation IBM Research, Tokyo Research Laboratory()
2nd Author's Name Hideki TAI
2nd Author's Affiliation IBM Research, Tokyo Research Laboratory
3rd Author's Name Gaku YAMAMOTO
3rd Author's Affiliation IBM Research, Tokyo Research Laboratory
Date 2006-07-14
Paper # DE2006-111
Volume (vol) vol.106
Number (no) 150
Page pp.pp.-
#Pages 6
Date of Issue