File systems - Motivation

Goal
‰ efficient and transparent access to shared files within a mobile environment
while maintaining data consistency
Problems
‰ limited resources of mobile computers (memory, CPU, ...)
‰ low bandwidth, variable bandwidth, temporary disconnection
Support for Mobility ‰ high heterogeneity of hardware and software components (no standard PC
architecture)
‰ wireless network resources and mobile computer are not very reliable
‰ standard file systems (e.g., NFS, network file system) are very inefficient,
almost unusable
‰ File systems Solutions
‰ replication of data (copying, cloning, caching)
‰ data collection in advance (hoarding, pre-fetching)

Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 1 Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 2

File systems - consistency problems File systems for limited connectivity I

THE big problem of distributed, loosely coupled systems Symmetry

‰ symmetric filesystems
‰ are all views on data the same? z clients run the meta manager code
‰ how and when should changes be propagated to what users? z all nodes undestand the disk structures
‰ asymmetric filesystems
Weak consistency z >= 1 dedicated meta managers that maintain file system and its associated disk structures
‰ many algorithms offering strong consistency (e.g., via atomic updates) ‰ client/server or peer-to-peer relations
‰ one file system or several file systems
cannot be used in mobile environments
‰ one namespace for files or several namespaces
‰ invalidation of data located in caches through a server is very problematic if z global namespace: all files organized under the same names and path names
the mobile computer is currently not connected to the network Transparency
‰ occasional inconsistencies have to be tolerated, but conflict resolution ‰ hide the mobility support, applications on mobile computers should not notice the
mobility
strategies must be applied afterwards to reach consistency again
‰ user should not notice additional mechanisms needed
Conflict detection Consistency model
‰ content independent: version numbering, time-stamps ‰ optimistic or pessimistic (conservative)
z pessimistic: clients access the latest data and data copies see the same order of updates
‰ content dependent: dependency graphs Caching and Pre-fetching
‰ single files, directories, subtrees, partitions, ...
‰ permanent or only at certain points in time

Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 3 Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 4
 File systems for limited connectivity II File systems - Coda I
Data management
‰ management of buffered data and copies of data Application transparent extensions of client and server
‰ request for updates, validity of data
‰ changes in the cache manager of a client
‰ detection of changes in data
Conflict solving ‰ applications use cache replicates of files
‰ application specific or general ‰ extensive, transparent collection of data in advance for possible future use
‰ errors (“Hoarding”)
Consistency
Several experimental systems exist
‰ Coda (Carnegie Mellon University), Little Work (University of Michigan), Ficus ‰ system keeps a record of changes in files and compares files after
(UCLA) etc. reconnection
‰ if different users have changed the same file a manual reintegration of the
Many systems use ideas from distributed file systems such as, e.g., AFS (Andrew file into the system is necessary
File System)
‰ optimistic approach, coarse grained (file size)
Andrew file system (AFS)
‰ uses a local cache (client-level caching) to improve filesystem performance
z reduce network and server workload and increase the performance of a distributed mobile client
computing environment
‰ location-independent filesystem: single shared namespace application cache server
‰ changed portions of a modified file are copied back to the file server after it is closed

Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 5 Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 6

File systems - Coda II File systems - Little Work

Hoarding States of a client ‰ Only changes in the cache manager of the client
‰ user can pre-determine a file list with ‰ Connection modes and use
priorities
‰ contents of the cache determined by
the list and LRU strategy (Last hoarding
strong Connected Partially Fetch only Disconnected
Recently Used)
connection Connected
‰ explicit pre-fetching possible weak Method normal delayed write optimistic abort at cache
‰ periodic updating disconnection connection to the server replication of files miss
Comparison of files write Network continuous continuous connection on none
‰ asynchronous, background disconnected requirements high bandwidth demand
‰ system weighs speed of updating connection bandwidth
against minimization of network Application office, WLAN packet radio cellular systems independent
traffic disconnection (e.g., GSM) with
Cache misses costs per call
emulating
=> fetch a file or request failed
‰ modeling of user patience: how long
can a user wait for data without an
error message?
‰ tradeoff: function of file size and
bandwidth

Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 7 Dr. Ka-Cheong Leung CSIS 7304 The Wireless Internet and Mobile Computing 8