Middleware based Data Replication providing Snapshot Isolation

1
Middleware based Data Replication providing
Snapshot Isolation

• Yi Lin
• Bettina Kemme
• Marta Patiño-Martínez
• Ricardo Jiménez-Peris
• June 15, 2005

2
Data Replication What,Why,How?
Without Replication
With Replication

• Problem
• Scalability,
• Fault Tolerance

Challenge Keep copies consistent
3
Data Replication challenge

• Replica Concurrency Control
• Keep copies consistent
• Isolate concurrent transactions

• General correctness criteria
• 1-copy-serializability

Replica control

• Our new criteria
• 1-copy-snapshot isolation

4
Implementing Replica Control

• Kernel-based approach
• Middleware-based approach
• Advantages
• Modular
• Do not need access to DB code
• Reusability
• Disadvantages
• No access to concurrency control information in
  the kernel

5
Related Work

• Some provide table-based concurrency control
  Middle-R, Conflict Aware, C-JDBC
• Some are primary/secondary approaches. Ganymed
• Updates must always be performed on primary copy
• Nearly all only look at 1-copy-serializability
  Conflict Aware, GlobData, Middle-R, C-JDBC,
  State Machine

primary
secondary
6
Concurrency Control serializability

• Serializability
• Highest transaction isolation level
• to what extend transactions interfere with each
  other
• The result is the same as executing them
  serially.
• Conflict read/write and write/write

T0
w(x)
T1
w(x)
r(x)
w(y)
T2
r(x)
w(x)
time
T0
w(x)
T1
w(x)
r(x)
w(y)
T2
r(x),w(x)
time
7
Concurrency Control Snapshot Isolation

• Snapshot Isolation (SI)
• Conflict only write/write
• Read from a snapshot of the committed data as of
  the time the transaction starts.
• First committer wins of 2 concurrent conflicting
  write transactions, one commits, the other aborts
• Similarity with optimistic concurrency control
• Very popular (Oracle, PostgreSQL, SQL Server 2005)

T0
w(x)
commit
T1
w(x)
r(x)
w(y)
abort
T2
r(x)
w(x)
time
8
Remainder of the Talk

• Define 1-copy-SI
• Protocols
• Centralized middleware
• Decentralized middleware
• Evaluation
• Conclusions

9
Our new correctness criteria 1-copy-SI

• 1-copy-SI The replicated system behaves as one
  database providing SI.
• Each DB provides SI.
• All read operations read from snapshots
  producible in one DB

...
...
1-copy-SI

SI
10
A Centralized Replica Control Protocol

• Challenge
• How to detect concurrent conflicting txns?
  ?Validation

succeed
fail
x
11
A Centralized Replica Control Protocol

• How to detect two transactions are conflicting?
• Writeset contains modified tuples and their
  corresponding primary keys.
• If two writesets share some primary keys, they
  conflict.
• Note Snapshot Isolation only cares about
  write/write conflicts.

Key1
T1
T2
12
A Centralized Replica Control Protocol

• How to detect two transactions are concurrent?

start0
end1
T0
end2
start1
T1
T2
start1
counter

• A counter for each database, increased upon
  committing a transaction
• Record start time and end time of transactions
• T0.end ? T1.start T1.end ? T0.start ? T0 and
  T1 not concurrent.

13

• 1-copy-SI
• more than just keeping data consistent

T1w1(x), val. succeed T2w2(y), val. succeed
r3(x,y),c3
r4(x,y),c4
14

• 1-copy-SI solution 1

T1w1(x), val. succeed T2w2(y), val. succeed
w1(x),c1
w1(x),c1
r3(x,y),c3
r4(x,y),c4
w2(y),c2
w2(y),c2
y
x
x
y
15

• 1-copy-SI solution 2

T1w1(x), val. succeed T2w2(y), val. succeed
r4(x,y),c4
x
y
w1(x),c1
w2(y),c2
x
y
x
y
r3(x,y),c3
r4(x,y),c4
w2(y),c2
w1(x),c1
x
y
r4(x,y),c4
y
x
x
y
16
Decentralized SIR Protocol

• Summary
• Validation? same as centralized approach
• Total order multicast ?all middleware components
  make the same decision

T1
T2
17
Experiment
No Replication
Decentralized
Table-based locking
5 sites, 100 update transactions,
18
Conclusions

• Definition of 1-Copy-SI
• Middleware based Replication Protocols providing
  1-Copy-SI
• Replication transparent to clients
• Connect via JDBC and submit operations as if
  non-replicated system
• Tuple based concurrency control

19
References

1. Conflict-Aware C. Amza, A. L. Cox, and W.
   Zwaenepoel. Conict-Aware Scheduling for Dynamic
   Content Applications. In USENIX Symp. on Internet
   Tech. and Sys., 2003.
2. C-JDBC E. Cecchet, J. Marguerite, and W.
   Zwaenepoel. C-JDBC Flexible database clustering
   middlerware, USENIX 2004
3. Ganymed C. Plattner and G. Alonso. Ganymed
   Scalable replication for transactional web
   applications. In Middleware, 2004.
4. GlobData L. Rodrigues, H. Miranda, R. Almeida,
   J. Martins, and P. Vicente. Strong Replication in
   the GlobData Middleware. In Workshop on
   Dependable Middleware-Based Systems, 2002.
5. Middle-R R. Jiménez-Peris, M. Patiño-Martínez,
   B. Kemme, and G. Alonso. Improving Scalability of
   Fault Tolerant Database Clusters. In ICDCS'02.
6. Postgres-R S. Wu and B. Kemme. Postges-R(SI)
   Combining replica control with concurrency
   control based on snapshot isolation. In ICDE,
   Tokoyo, Japan, 2005.
7. State Machine F. Pedone, R. Guerraoui, and A.
   Schiper. The Database State Machine Approach.
   Distributed and Parallel Databases, 1471-98,
   2003.

20
Writeset

• Containing modified tuples and their primary keys
• Implemented by triggers or log sniffing
• Currently not provided by DB standard interfaces
• PostgreSQL is currently implementing such
  interface