MYE017 Distributed Systems. Kostas Magoutis

Transcription

1 MYE017 Distributed Systems Kostas Magoutis

2 Data-centric Consistency Models The general organization of a logical data store, physically distributed and replicated across multiple processes.

3 Strict Consistency Any read on a data item x returns the value of the most recent write to x Behavior of two processes operating on the same data item. The horizontal axis is time.

4 Weaker Consistency a 1 2 x x Behavior of two processes operating on the same data item. The horizontal axis is time.

5 A data store is sequentially consistent (SC) when: The result of any execution is the same as if the (read and write) operations by all processes on the data store were executed in some sequential order on a single copy of the data store the operations of each individual process in this sequence appear in the order specified by its program.

6 More formaly E i : Sequence of read or write operations executed by process P i over data store S E.g. E 3 = R 3 (x)br 3 (x)a History H : sequence of op executions over hypothetical centralized data store S H is an interleaving of E i i=1,.., n All acceptable histories H must respect The order of operations in individual executions Data coherence (read last value written)

7 H : W 2 (x)br 3 (x)b R 4 (x)b W 1 (x)a R 3 (x)a R 4 (x)a (a) A sequentially consistent data store. (b) A data store that is not sequentially consistent.

8 $1 1 2 $2 x y x y Actual execution: B A Naïve replication scheme: (a) Read/write local replica (b) For writes, return before updating remote replica (c) If local replica fails, use another one Example from Coulouris, Dollimore, Kindberg, Blair, 5th Edition, Addison Wesley, May All rights reserved

9 $1 1 2 $2 x y x y Actual execution: B A Naïve replication scheme: (a) Read/write local replica (b) For writes, return before updating remote replica (c) If local replica fails, use another one Example from Coulouris, Dollimore, Kindberg, Blair, 5th Edition, Addison Wesley, May All rights reserved

10 Three concurrently-executing processes. How many (and which) histories are sequentially consistent?

11 Examples of valid execution sequences. The vertical axis is time.

12 720 (6!) histories possible, not taking program order into account Start with x=1, consider 120 (5!) possible sub-histories Only 90 out of 720 respect program order 2 6 = 64 signatures possible, however not all valid Are signatures , valid according to SC? 90 histories that respect program order produce <64 valid outputs Sequential consistency forms a contract between processes and the distributed data store

13 Linearizability A data store is linearizable when: The result of any execution is the same as if the (read and write) operations by all processes on the data store were executed in some sequential order on a single copy of the data store the operations of each individual process appear in this sequence in the order specified by its program. Additionally, If tsop1(x) < tsop2(y) then OP1(x) must precede OP2(y) in this sequence