490dpPrelude I: Group MembershipPrelude II: Team Presentations

Tuples

Robert Grimm

Group Membership

• c490dp-a– awong– boriss– dancheah– ishafe– sfanous

• c490dp-b– bret– cyhung– davidjc– kphilbri– muliawan

Try to edit /projects/instr/01wi/cse490dp/[group name]/foo

Team Presentations

• Problem(s)– The problem(s) your application is

addressing• Features

– The functionality of the application• Assumptions

– Premises about the application domain

Team Presentations

• Components– The structure of the application

• Unknowns– Things you aren’t sure about– The plan for resolving them

• Evaluation– The plan for showing that the application

• Solves the problem(s)• Works as advertised

Tuples

Parallel Computing

• How to coordinate processes?– Use message passing

• Point-to-point or group communication• Synchronization

– Problem• Relatively tight coupling

– Can we do better?

Gelernter Did Better

• Linda [Carriero & Gelernter 86]

– Tuple• Ordered collection of typed fields• <“P”, 5, false>

– Tuple space• Globally shared collection of tuples

Operations

• Three basic operations– out: Write a tuple– in: Consume a tuple– read: Read a tuple

• Unblocking and blocking versions– Blocking: Wait for result

Template Matching

• in and read use templates– Ordered collection of formals and values

• <“P”, int i, bool b>– A template matches a tuple if

• They have the same size• Corresponding values are equal• Corresponding formals and values agree

– Multiple matches: pick any of them

A Simple Application Example

• Main program– Write out job descriptions– Start helper programs– Collect results

• Helper programs– Take job description– Perform computation– Write out result

Advantages of Tuple Spaces

• Destination uncoupling– Fully anonymous communications

• Space uncoupling– Associative access independent

of local process• Time uncoupling

– Tuples stored independentof creating process

Distributed Computing

• Coordination– Similar problems as for parallel computing– Considerably more dynamic environment

• More generally: Data management– How to represent data?– How to store data?– How to exchange data?

Advantages of Tuple Spaces

• Destination uncoupling• Space uncoupling• Time uncoupling

• Idea: Adopt tuple spaces for distributed computing

Systems

• Three systems– JavaSpaces– T Spaces– Structured I/O in one.world

• In common– Data represented as tuples

JavaSpaces

• Tuple space in / for Java– Typing– Leases– Transactions– Architecture– Potpourri

Typed Tuples

• Convention in Linda– First field is a string denoting the type

• Strongly typed language: Java– Type provided by class– Unordered collection of

named and typed fields• Public instance fields

–References, but not primitive types

Template Matching

• Template is a tuple– Fields may be null

• A template matches a tuple if– The tuple has a subtype of the template– Corresponding values are equal

• Fields are serialized separately– Possible to compare serialized form

Leases

• Problem– How to reclaim obsolete tuples?

• Solution– Limit their lifetime through leases

• Acquire• Renew• Cancel

Transactions

• Problem– How to group operations into one unit?

• Solution: Transactions– Either all operations complete (commit)– Or no operation complete (abort)– Operations become visible only at commit– Affected tuples are locked down

• Serializability

Architecture

• Tuple space server– Provides storage– Performs operations

• Client– Linked in with application– Communicates with server

Potpourri

• notify operation– Triggered when matching tuple is written

• Unblocking versions– readIfExists, takeIfExists

• Blocking versions– read, take

• Implementation may be persistent

T Spaces

• How does T Spaces differ from JavaSpaces?

T Spaces

• Hierarchy of tuple spaces– More flexible structuring

• Support for queries– Data management

• Support for access control– Security

• Extensible implementation– Application-specific functionality

Issues

• What’s wrong with JavaSpaces or T Spaces?

Issues

• Single point of failure– Tuples stored on centralized server

• Relatively heavy-weight– Client to server to client– Real-time communications

• Multimedia

How to Address These Issues?

• Localize storage– Add support for replication

• Offer light-weight communications

Structured I/O

• Tuple-based• Storage

– Local within environment hierarchy• Communication channels

– Unreliable• On top of UDP

– Reliable• On top of TCP

Detour: Environments

• Containers for– Stored tuples– Application components– Other environments

• Combination of file system directories and nested processes

• Local root functions as the kernel

Tuples in one.world

• Statically typed tuples: Subclasses of Tuple– All instance fields must be public– Fields may be of any type– Default fields: id, metaData

• Dynamically typed tuples: DynamicTuple– Field declared and typed dynamically

• Tuples may be nested

Operations

• Basic operations– Communications and storage– put, read, listen

• Extended operations– Storage only– May be transactional (not implemented)– put, read, listen, query, delete

Queries

• Used by read, listen, and query operations• Empty query: always true• Unary query: Negation• Binary query: Conjunction, disjunction• Comparisons

– Including ==, <, >, beginsWith, …– hasType, hasSubType, hasDeclaredType, hasField

Support for Replication

• Replication 101– Log modifying operations– Propagate operations to other nodes

• How to name data?• How to layer replication on top of storage?

The Use of IDs

• Modifying operations use ID as primary key– put operation

• ID differentiates between–Adding a tuple

»No tuple with same ID in store–Overwriting a tuple

»Tuple with same ID in store– delete operation: By ID only

Why No take Operation?

• Result only known at end of operation– Harder to log

• Can still be expressed– Transactional read & delete

Discussion

• Questions?