Modern Concurrency Abstractions for C# by Nick Benton, Luca Cardelli & C´EDRIC FOURNET Microsoft Research

Outline Introduction Polyphonic C# overview Specification Programming in Polyphonic C# Related work Future work Conclusion

Introduction How is concurrency implemented ? Language feature Libraries Provides features like – memory management & exceptions Advantages of having them as language features Produce better code Warn programmers of potential & actual problems

They made concurrency as a language feature Monitors - more asynchronous Shift in focus – shared memory to event- or message- oriented concurrency Programming constructs handling asynchronous communication Polyphonic C#

Polyphonic C# Overview Extension to C# - new asynchronous concurrency constructs based on join calculus New constructs – Asynchronous methods Chords Asynchronous methods – guaranteed to complete essentially immediately Never returns a result async keyword is used

Chord – consists of a header and a body Header - set of method declarations Body is only executed once all the methods in the header have been called Method calls are implicitly queued up until/unless there is a matching chord If buff is instance of Buffer then call to buff.Get() has two possibilities If there has previously been an unmatched call to buff.Put(s) If there are no previous unmatched calls to buff.Put(s)

Conversely - call to the asynchronous method buff.Put(s) has two possibilities If there has previously been an unmatched call to buff.Get() If there are no pending calls to buff.Get() Exactly which pairs of calls are matched up is unspecified Difficulties In which thread the body runs ? To which method call the final value is returned ?

Specification If return-type is async then the formal parameter list formals may not contain any ref or out parameter modifier At most one method-header may have a non-async return- type. If the chord has a method-header with return-type type, then body may use return statements with type expressions, otherwise body may use empty return statements. All the formals appearing in method-headers must have distinct identifiers. Two method-headers may not have both the same member- name and the same argument type signature.

The method-headers must either all declare instance methods or all declare static methods All method-headers with the same member-name and argument type signature must have the same return- type and identical sets of attributes and modifiers Typing issues – async is a subtype of void an async method may override a void one, a void delegate may be created from an async method, and an async method may implement a void method in an interface

Programming Polyphonic C# A simple example Put Chord Get Chord

Reader writer lock Problems with the above code Fairness problem – starvation(writer)

Reader Writer lock

Combining Asynchronous Messages

Active Objects

Implementation Translation from Polyphonic C# to plain C# Converting state chords to state automata synchronization state consists of the pending calls threads for regular methods messages for asynchronous methods synchronization depends on – presence or absence of pending calls number of calls actual parameters

Performance issues – Complete asynchronous chord is always expensive asynchronous method synchronous method All asynchronous messages are present Otherwise Translation – modularize the translated code by introducing auxiliary classes for queues and bitmasks Queues – class that represents message queues

Related work The work is based on join calculus Join java follows similar approach of using join calculus Synchronization barriers in dataflow languages modeled in Petrinets are similar to chords.

Future Work Scope for optimizing the implementation Lock optimization - There are situations when we could safely ‘fuse’ successive critical sections protected by the same lock. Queue optimization - at most one pending call on a particular object could be compiled using private fields instead of queues Thread optimization - Purely asynchronous chords can often be compiled to run in the invoking thread, rather than a new one

Conclusion Designed and implemented a join calculus-based extension of C# that is simple, expressive, and efficient. Writing correct concurrent programs is considerably less difficult in Polyphonic C# than in ordinary C# Implementation is constrained by the underlying threads-and-locks model

QUESTIONS