230 likes | 412 Views
Extending Tuplespaces for Coordination in Interactive Workspaces. https://graphics.stanford.edu/papers/eheap-jss/. Enrica Dente Digital Enterprise Research Institute enrica.dente@deri.org 2004/07/26. Objective. To understand: What is Tuplespace computing?
E N D
Extending Tuplespaces for Coordination in Interactive Workspaces https://graphics.stanford.edu/papers/eheap-jss/ Enrica Dente Digital Enterprise Research Institute enrica.dente@deri.org 2004/07/26
Objective • To understand: • What is Tuplespace computing? • Event Heap Implementation for Interactive Workspaces • Relevance to WSMO
Overview • Interactive Workspaces Requirements • Tuplespace Model • Event Heap: Tuplespace Model + Extensions • Summary • Relevance to Semantic Web Services • Deri Triplespace Computing Projects
What is an Interactive Workspace? • Interaction with or ensembles of applications/computers • No clear coordination model at the application level • IRoom prototype “It is an ubiquitous computing environment where groups come together to collaborate on solving problems”.
What is an Interactive Workspace?(cont.) • It is a dynamic environment for integration and coordination • It contains : • Embedded devices: • Permanent computational and I/O resources embedded into the environment • Mobile devices: • Allows portable devices brought into the space to be used in conjunction with embedded facilities (i.e ongoing) • Interoperability • Autonomy (of participating organizations) • Trust, privacy and security.
Interactive Workspace Factors (learning from iRoom prototype) • Technology factors: (i.e coordination, integration, dynamism) • Changing environment (i.e short and long term change) • Heterogeneity (i.e rapid integration of heterogeneous hardware, software and software platforms) • Human factors: (i.e problem solving, collaboration) • Bounded Environment (i.e only within IW’s physical space ) • Human Centred Interaction (i.e can reconfigure tools depending on problems arising) • Human Level Performance needs (i.e can set constraints).
Interactive Workspace Requirements • Must facilitate common IW Usages: • Moving data between devices • Controlling devices and applications from other devices • Coordinating applications. • Must meet these requirements: • Heterogeneity of devices in the workspace, and new devices being added over time • Easy integration of legacy devices, using whatever communication mechanisms they provide. • Robustness and availability despite software/hardware failure • Portability across installations.
Interactive Workspace Desired Properties • Limited Temporal Decoupling • Data relevance until needed • Communication continuity • Referential Decoupling • Limited inter-dependency, still interaction possible • Location independence • Extensibility • Snooping • Interposability • Stream transformation • Expressiveness • Coordination and distribution patterns
Interactive Workspace Desired Properties (cont.) • Simple and Portable client API • Reduced number of API calls and amount of code • Easy Debugging • Easy to figure out and debug interactions • Perceptual Instantaneity • Coordinations and actions instantaneous to user • Latencies minimized (i.e 50 ms for round trip) • Failure Tolerance and Recovery • Mechanisms for coping with failures • Application Portability • Coordination infrastructure independent from IW. • Scalability to Workspace-sized Traffic loads • Load limited by number of entities connected and rate of updates
Properties Supported by Coordination Infrastructures • In Pub-Sub and MOM: • Application portability • Query persistence and notification via subscription • In RMI/RPC: • Perceptual Instantaneity only • In Tuplespace Model and Pub-Sub: • Referential Decoupling • Simple and Portable Client API • Perceptual Instantaneity • Extensibility (limited support) • Easy debugging (limited support) • Further Reading: • InfoBus system • Eugster et al. overview • Siena system • Tuplespace Modelhas most properties needed for Interactive Workspaces. • In Tuplespace Model also: • Limited Temporal Decoupling • Expressiveness • Failure Tolerance and Recovery (limited support)
Linda Tuplespace Model*for parallel computing • Goal: add concurrent programming capability to sequential programming • Need strong decoupling of processes in space, time and reference • Solution: place data as tuples in the space for process coordination • Tuple: • object named by tag and key (similar to content addressable memory) • symbolic name + an ordered list of typed values • equally accessible to all processes but bound to none • Tuplespace: • “shared memory" for accessing, modifying and deleting tuples • data fields in space matched with elements in requesting template • must have same number of fields, same types of fields in same order, (possibly identical values as in the template) • retrieval made blocking when there is no match • actuals assigned to formals when match found. *Carriero and Gelernter, 1986 *Carriero and Gelernter, 1986 *Carriero and Gelernter, 1986 *Carriero and Gelernter, 1986
In () Receiver 1 Read () Out () Receiver 2 Sender Linda Tuplespace Model*for parallel computing (cont.) • 6 funtions: • out(<tuple>) -inserts a tuple in the space. It never blocks a tuple. • rd(<template(tag, list)>) - retrieves all tuples which match a given template from the TS. It can be both blocking and non blocking. • in<template(tag, list)> - retrieves and removes all tuples which match a given template from the space. This function blocks. • eval(<function-tuple>) - creates an active tuple and evaluate its entries. The results are stored as a passive tuple(i.e data) in the space. Tuplespace Figure 1 Tuplespace Model for parallel programs *Carriero and Gelernter, 1986
Linda Tuplespace Model*for parallel computing (cont.) Process 1: Ping () { loop { out("PriceOfBook", "100") in("PriceOfBook", "200") } } Process 2: Pong () { loop { in("PriceOfBook", "100") out("PriceOfBook", "200") } }
Tuplespace Model for Interactive Workspaces • Routing and Content Addressing provides reference decoupling • Persistence providestemporal decoupling • Limited Transparency providesextensibility and debugging • Centralization providestime and reference decoupling • Simple and general infrastructure API: 6 functions in any programming language.
Event Heap:Tuplespace Model + Extensions • Limitations: • For processes designed to work together (parallel programming) • Extensibility and application portability not supported • Push not supported.No guarantee for receipt of events • Benefits • Integration and Extensibility • Application Portability • Easy debugging • Anonymous Coordination • Interposability • Snooping • Expressiveness • Failure Tolerance + Recovery. • Extensions: • Self-Describing Tuples • Flexible Typing • Typed Tuples • Tuple Sequencing • Tuple Expiration • Standard Routing fields • Query registration + notification • At Most Once Ordering (FIFO) • Modular Restartability.
Extensions Supported in Tuplespace Implementations • GigaSpaces and TSpaces • Self-Describing Tuples • Flexible Typing (limited support) • Tuple Expiration • Query Registration • L2mbo • Flexible Typing • Tuple Expiration • LIME • Registration • At Most Once Ordering • Modular Restartability • Standard Routing Fields (limited support) • Known implementations support few extensions
Event Heap ExtensionsSource: http://graphics.stanford.edu/papers/eheap3/ • Event standard fields • User fields (EventType, TimeToLive) • Routing fields (target and source fields) • Internal user fields (SessionID, SequenceNum, EventHeapVersion) • Field structure: • Type • Name • Post Value • Template Value • Permitted values: • Actual • Formal • Virtual • Auto-set, auto-set overrideable Figure 2: Example operation showing placed events, and using template events for matching
Summary • IWsrequire reusable, portable and robust software infrastructure • Need uniform, useful, abstraction, that automatically manages shared data and state. • Tuplespace Model, RMI/RPC, Pub-Sub, MOM don’t have all properties required • Tuplespace Model best suited to Interactive Workspaces: • Supports limited temporal decoupling and expressiveness • least amount of extensions needed, compared to the other models (i.e portability) • Event Heap: Tuplespace Model + Extensions address issue of portability • General applicability to ubiquitous computing and to the web to be demonstrated.
Open Issues • Centralization reduces scalability (not suitable for the web) • Extensions support heterogeneity but sacrifice performance • Improving performance (more events per second at latency of less than 50 ms for round trip) • Content-based matching with streaming of high bandwidth data (e.g sound or video) not suitable • Extensibility and easy debugging versus security and privacy (i.e transparency of data communication) • More coordination and communication types • Tuple types and fields names collisions (i.e if they have the same permitted values) • Each value expressed as an RDF type? (i.e semantic typing) • Use URI to give address to tuple and group related triples?
Relevance toSemantic Web Services • Web Services are about coordination and integration: • Extensibility and application portability • Failure tolerance and recovery • Easy debugging and communication (e.g receipt of all events) • Performance and scalability • Need a set of policies (i.e ontologies) from specialised middleware: • Routing compatibility • Ordering • Limited Data Persistence, Query Persistence and registration • Communication transparency (address security and privacy issues) • Distributed Infrastructure (address performance and scalability issues) • Modular restartability • Spaces can store triples, triples can store data on spaces.
Deri Triplespace Computing Projects • D21: Triplespace computing and WSMX • State of the art and implementation issues • Relationship with WSMX Event Manager? • Relationship with WSMX Invoker? • Develop sets of components outside WSMX? • FIT-IT project (in collaboration with Eva Kuehn from Tecco) • Corso extensions to publish and retrieve triples • Proposal and work packages in draft status • Architecture in draft status.
References • [Fense,l 2004] DERI Research Report 2004-05-31, Triple-based Computing, http://www.wsmo.org/2004/tp-computing/ , 2004 • [Gelernter, 1992] D. Gelernter, Mirrorworlds, Oxford University Press, 1992 • [Johanson, 2002] Brad Johanson and Armando Fox The Event Heap: A Coordination Infrastructure for Interactive Workspaceshttp://graphics.stanford.edu/papers/eheap3/ , 2002 • [Fenwick, 1996] James B. Fenwick Jr., Lori L. Pollock. Issues and Experiences in Implementing a Distributed Tuplespace. Technical Report TR 9706, University of Delaware, 1996. http://www.eecis.udel.edu/~pollock/papers/spe_deli.ps • [Kühn, 2001] Kühn Eva, Virtual Shared Memory for Distributed Architecture, Nova Science Publishers, 2001