Presentation is loading. Please wait.

Presentation is loading. Please wait.

Disconnected Operation in the Coda File System

Similar presentations


Presentation on theme: "Disconnected Operation in the Coda File System"— Presentation transcript:

1 Disconnected Operation in the Coda File System
James J. Kistler M. Satyanarayanan

2 History Descendent of AFS, also a CMU project
In development since 1987 Support is included in the official Linux kernel

3 Motivation Distributed file systems are not immune to failures
Laptop computers are popular

4 Key Points Caching file data can be used to provide high availability
High availability allows disconnected operation The Coda file system provides these features

5 Design Goals Scalability Voluntary vs. Involuntary Disconnections
First vs. Second Class Replication Optimistic vs. Pessimistic Replica Control

6 Scalability Experience with AFS Whole-file caching
Functionality at clients, rather than servers

7 Voluntary vs. Involuntary Disconnections
Laptops Types of disconnection are not functionally different

8 First vs. Second Class Replication
Why have server replicas at all? Server replicas (first class) High quality Additional cost Client replicas (second class) Generally inferior Very little cost

9 Optimistic vs. Pessimistic Replica Control
Pessimistic control Avoids conflicting operations Clients acquire shared or exclusive control of objects Optimistic control Reads and writes everywhere! Detects and resolves conflicts after they occur

10 Venus States

11 Venus States: Hoarding
Client is connected to server Caches information in anticipation of disconnection Factors complicate implementation Prioritized cache management Hoard walking

12 Hoarding: Prioritized Cache Management
Hoard database Explicit prioritization, generated by user Contains pathnames and priorities corresponding to file system objects Recent reference history Implicit prioritization Updated continuously as objects are referenced

13 Hoarding: Hoard Walking
Equilibrium Two phases: Reevaluate HDB entries to reflect activity by other clients Reevaluate priorities and restore equilibrium

14 Performance Duration of reintegration Cache size
Likelihood of conflicts

15 Cache Size

16 Likelihood of Conflicts
Data based on AFS server usage Tracked who made modifications Measured time between modifications Promising results: 99% of all modifications were made by the previous writer

17 Likelihood of Conflicts

18 Recap Coda provides disconnected operation by aggressively caching file data

19 My Evaluation Resource exhaustion Likelihood of conflicts

20 More Info


Download ppt "Disconnected Operation in the Coda File System"

Similar presentations


Ads by Google