The dangers of replication and a solution

Abstract

Update anywhere-anytime-anyway transactional replication has unstable behavior as the workload scales up: a ten-fold increase in nodes and traflc gives a thousand fold increase in deadlocks or reconciliations.Master copy replication (primary copyj schemes reduce this problem.A simple analytic model demonstrates these results. A new two-tier replication algorithm is proposed that allows mobile (disconnected) applications to propose tentative update transactions that are later applied to a master copy. Commutative update transactions avoid the instability of other replication schemes.

Pronto: High availability for standard off-the-shelf databases

“…Initially, the client chooses one database server (lines 7-8), tries to start the transaction on this server (line 9), and waits for an • Executing a transactional job (lines [15][16][17][18][19][20][21][22][23][24][25]. The client sends the transaction requests, resulting from the execution of the transactional job, to the primary.…”

“…Gray et al [23] have classified database replication into eager and lazy. Eager replication, the category Pronto falls into, ensures strong consistency.…”

“…Lazy replication increases performance by allowing replicas to diverge, possibly exposing clients to inconsistent states of the database. Eager replication can be based on the update everywhere or on the primary-backup approach [23]. Update everywhere allows any copy of the database to be updated.…”

Increasing the Expressiveness of Analytical Performance Models for Replicated Databases

“…Additionally, the model of transaction processing in 2RC follows the primary copy approach [30], since it has been judged advantageous over other replica control concepts [15,18] and is implemented in commercial systems like Sybase and Oracle. In 2RC, asynchronous update propagation to the secondary copies is modeled, i.e.…”

“…If 100% of the data items are replicated, the throughput grows to about 100 TPS as the number of sites is increased to 30. Larger systems do not achieve a significantly higher throughput because without relaxed coherency high replication in large systems causes considerable update propagation overhead which hinders scalability [15]. Reducing replication (r 1 < 1) gradually improves scalability.…”