Synchronization. Physical Clocks Solar Physical Clocks Cesium Clocks International Atomic Time Universal Coordinate Time (UTC) Clock Synchronization Algorithms.

Synchronization

Physical Clocks

• Solar Physical Clocks

• Cesium Clocks

• International Atomic Time

• Universal Coordinate Time (UTC)

• Clock Synchronization Algorithms– One machine is connected to WWV– No machine is receiving WWV time

Election Algorithms

• Many distributed systems require one process to act as coordinator

• Assume:– All processes are the same– Each process has a process number– Each process knows the process number of

every other process

The Bully Algorithm

• When a process P notices the coordinator is no longer responding it initiates and election– P sends ELECTION to every process with a

higher number– If no one responds P wins and becomes

coordinator– If one of the higher-numbered processes

responds then it takes over and P is done

The bully algorithm

Ring Algorithm

• The process noticing that the coordinator is not functioning builds an ELECTION message containing its process number and sends it its successor

• When the message gets back to the process that started it, its type is changed to COORDINATOR and sent around the ring

• The process with the largest number is elected.

Ring Algorithm

Atomic Transactions

• A transaction is either committed completely or nothing is done (all-or-nothing)

• Sample tape based application

Atomic Transaction

• System consists of some number of independent processes

• Each process can fail at random

• Assume communication errors are handled by underlying software

Storage types

• Volatile RAM memory

• Disk storage

• Stable storage– Implemented using two or more disks– In case a failure, the second disk is used for

recovery

Stable storage

Transaction primitives

• BEGIN TRASACTION

• END TRANSACTION

• ABORT

• READ

• WRITE

Example Transaction

Properties of Transactions

• Serializability

Concurrent transactions do not interfere with each other

• Atomicity• To the outside world, the transaction happens

indivisibly

• Permanence• Once a transaction commits, the changes are

permanent

Nested Transactions

• The transactions can be contain sub-transactions

• Either all sub-transactions should commit or abort

• Each sub-transaction is given a private copy of data

Implementation

• Private Workspace– The copying is expensive– Optimizations for read and write are available

• No copy if no write is going on• Only copy the block list of files and update

modified blocks

Private Workspace

Writeahead Log

• A list called an intension list or LOG is created with TransID, BlockNo, OldVal, NewVal fields

• After the log has been successfully written, the changes are made to the file

• In case of ABORT, the operation are reversed from last to first

Writeahead LOG

Two Phase Commit

• Send each subordinate a message

• A subordinate responds when it is ready to commit

• If all subordinates are ready then commit else abort

Two Phase Commit

Concurrency Control

• Locking– A process can lock a file before a transaction– Smaller piece of data can be lock for

efficiency (Granularity of locking)– Two phase locking (growing and shrinking)

can be used to avoid deadlocks and cascaded aborts

Optimistic concurrency Control

• Perform any transaction without considering other processes

• When committing the transaction, check if the files have been changed or not

• It is DeadLock Free

• In case of failure the transaction should re-run

Time stamp

• Assign a time stamp to each transaction at the moment it begins

• On committing check the orders– If orders true then commit – Else abort

Time stamp

Questions?