1. Adaptable Consistency Control for Distributed File Systems Simon Cuce simon.cuce@csse.monash.edu.au Monash University Dept. of Computer Science and Software Engineering

2. Simon CuceSchool of Computer Science and Software Engineering Monash University Background DFS provide additional subsystems for replication and one- copy equivalence DFS provide additional subsystems for replication and one- copy equivalence  Replication  Availability and Performance  One-Copy Equivalence  Concurrency Control  Consistency Maintenance Current DFS implementations Current DFS implementations  Mostly cater for one scenario  Fail to work effectively and efficiently when underlying assumptions change  Inflexible to change  Difficult to adjust and modify

3. Simon CuceSchool of Computer Science and Software Engineering Monash University Issues DFS topologies include different DFS topologies include different  Hardware Components  Network Types  Applications and Users Providing an effective and efficient level of service is impossible Providing an effective and efficient level of service is impossible  Each has their own expected level of consistency requirements  Difficult to anticipate future scenarios  A universal approach is not feasible Improve the process used for consistency control mechanisms Improve the process used for consistency control mechanisms  Such that extending is simplified

4. Simon CuceSchool of Computer Science and Software Engineering Monash University The Solution Abstract the consistency control mechanism out of the application and/or OS and re-implement within a suitable framework Abstract the consistency control mechanism out of the application and/or OS and re-implement within a suitable framework Provide numerous consistency control mechanisms to cater for different scenarios Provide numerous consistency control mechanisms to cater for different scenarios Improve the mechanism of consistency control creation and management via a component-oriented architecture Improve the mechanism of consistency control creation and management via a component-oriented architecture Create a framework that allows for the supporting of existing and future heterogenous systems by promoting openness Create a framework that allows for the supporting of existing and future heterogenous systems by promoting openness Provide enough flexibility that alternative approaches can be catered for Provide enough flexibility that alternative approaches can be catered for

5. Simon CuceSchool of Computer Science and Software Engineering Monash University GLOMAR GLOMAR is that framework GLOMAR is that framework  Provide a mechanism to encapsulate alternative consistency control mechanisms under the one distributed file system  Provide support services to control the implementation of these encapsulated consistency control mechanisms  Provide a methodology for the simplified creation of consistency control mechanisms Two main design elements Two main design elements  The Relationship Component (RC)  The component that encapsulates the consistency control functionality and scope information  The System Support Components (SSC)  Manages the RCs  Implements a RC when appropriate  Provides support services for consumption by the RC

6. Simon CuceSchool of Computer Science and Software Engineering Monash University Relationship Component Conceptually consists of 3 elements Conceptually consists of 3 elements Consistency Model, Relationship Scope and Clone List

7. Simon CuceSchool of Computer Science and Software Engineering Monash University Relationship Component The Consistency Model The Consistency Model  Encapsulate the algorithms used to maintain consistency  Based on file system primitives (read, write, etc) The Clone List The Clone List  Defines what replicas this RC should govern The Relationship Scope The Relationship Scope  Used to define what conditions this RC is suited.  User defined  Resulted based on heuristic generated from the System Grader The Relationship Component The Relationship Component  A bound collection of a CM, a RS and a CL  And other metadata (threading, instantiation and life cycle)

8. Simon CuceSchool of Computer Science and Software Engineering Monash University System Support Components Handles the management of RCs. Handles the management of RCs. Sub components Sub components  The Interceptor  Intercept IO requests for further processing  The System Grader  Provides and constantly updates the heuristic information  The Remote Operation Interface  Handles requests from remote machines  The Clone Distribution Monitor  Provides replica name resolutions  The Relationship Component Repository  Manages and selects RC  Implements Relationship Component Processing (RCP)  The Service Manager  Allows installation of developer created services  The Executive  Mangers all events and sub systems

9. Simon CuceSchool of Computer Science and Software Engineering Monash University System Support Components

10. Simon CuceSchool of Computer Science and Software Engineering Monash University GLOMAR Implementation Written in Microsoft.NET Written in Microsoft.NET  ~6000 lines of code  Supports component programming  Multiple language support (for legacy CC implementations)  Support for new technologies (e.g. SOAP) Runs as a Windows Service Runs as a Windows Service Administration Console Administration Console  GUI used to detail GLOMAR elements Number of heuristic libraries supplied Number of heuristic libraries supplied  Connectivity, Processor, Network, Memory

11. Simon CuceSchool of Computer Science and Software Engineering Monash University RC Implementations Document Editing Application Document Editing Application  Illustrate the feasibility of GLOMAR  Support for disconnection operations  Implementation  A “notepad like” application that persists per key stroke  Two RCs ROWA (Pessimistic)ROWA (Pessimistic) Get Latest (Optimistic)Get Latest (Optimistic)  Uses SOAP and web services MS Outlook 2000 RC MS Outlook 2000 RC  Implement a different consistency model for different folders  Notes, DraftMail, SentMail, Inbox, Contacts, Calendar and Tasks

12. Simon CuceSchool of Computer Science and Software Engineering Monash University RC Implementations Twin Transaction Model Twin Transaction Model  PhD by Aamir Rasheed  Created by an external party  Illustrates the implementation of an advanced approach to concurrency control within a DFS  Transactional base