Fichas
Lecture 6 - file systems
Fichas del conjunto (16)
- 1File services: Permanent storage - Named set of objects called files - They come into existence by explicit creation - They are immune to temporary system failures - They persist until explicitly destroyed Hardware background - They are secondary storage devices - They have high capacity - Slow access time (sequential or random access) (RAM disk: permanent storage methods for volatile storage)File Systems:
Functions
- Part of the OS for permanent data & program storage
- Concerns with organisation, storage, retrieval, naming, sharing, protection of files
- Basic operations: create, delete, read, write
Types:
- Conventional file system: uniprocessor based, time sharing OS
- Distributed file system - multiplicity and autonomy of clients and servers - 2Single-user file system issues: Organisation of files: - Directory / subdirectory of tree - Problem: directory cycles Naming : - File extensions can identify file content Programming interface (access mode): - Mapping into user address space - eg: Multics - Set of file operations - eg: Unix - Atomic transactions (changes not permanent until committed)Physical storage:
- File mapping on the storage media
- block sizes vs fragmentation
Integrity
- File consistency protection against power, hardware, and software failures - 3Multi-user file system issues: Concurrency control - Simultaneous access by a number of users - Synchronisation polices - operational conflicts need to be resolved run-time - Serialisability - the results of concurrent operations should be the same as if they were applied one after the other - Deadlock avoidance, detection, resolution - if users can lock items, continuous operation needs to be maintainedSecurity
- User identification authentication
- User grouping
- Access privileges for file operations - 4Distributed file system issues - User mobility: full access from different terminal (workstations etc.) - Performance degradation: remote acess overhead - Scalability - scope for system growth before saturation - Availability - fault tolerance, ie: single point errors should not bring the system down - Heterogeneity - accomodate a variety of workstations - Security - increased security against network intrusion - Integrity - increased protection against network problemsFile services
- 5—File system modules:
- Client module
- integrate the diff modules in a single application programming interface (API)
- Directory module: relates file names to file IDs
- File module: related file IDs to particular files
- Access control module: checks permission for operation requested
- File access module: reads and write file data or attributes
- Block module - accesses and allocates disk blocks
- Device module - disk I/O and buffering - 6Services - Directory service - mapping between text names and file references / Unique file identifiers/ UFIDs - Flat file service - data and attributes handling (eg: file size, access time, owner) - Block service - allocation of disk blocks, operations for storing and retrieving data - Service cooperation problems - space leaks (lost block/files)Directory services:
- Operations on file names:
Add_Name, Delete_Name, Rename
Look_Up, Get_Names
Tree structures directories - Hierarchical file systems:
- Files are terminal nodes, directories are interior nodes
- Pathname: List of directories through the tree, from root to file
- Files can be in multiple directories
Eg: symbolic links in UNIX
Access control:
- Basis of access control - client identification or user supplied capabilities (eg: cryptographic keys)
- File ownership - the owner of a file has total control over the file
- Access modes, eg: Owner, group, world: read, Write, Edit - 7Flat file service : - Flat file system - records contain no structural relationships (eg: no directory) - Operations on files - Modifying file content - write, truncate - Modifying file existence - create, delete - Observer type - read, length of file - Immutable files - cannot be modified after creation, to edit a file a new version of it has to be created - Idempotent operations - duplicated operations cause no harm - may degrade performance, eg: clients repeat on timeouts with overloaded servers Operations on file attributes - eg: creation / modification time, file type, etcFile system organisation
- Block service: separates implementation of the file service from disk specific optimisation and hardware concerns
- Operations: Allocate_Block, Free_Block, Get_Block, Put_Block
- File groups: collection of files mounted on a server machine
- Files cannot change group membership - 8—
- 9Stateless servers - Do not store state information on behalf of clients, eg: no read/write pointer stored Advantages - Fault tolerance: server crashes: state info not lost client crashes: unused data not stored in server - No space wasted on tables - No limit on the number of open files Disadvantages: - Long messages - A message needs to contain all relevant info - Slower processing - parameters may need to be recomputed several times - Difficult to implement idempotency - No file licking for exclusive access
- 10Caching: - Cache: holds small amount of replicated for faster access - Basic terms: process communicates with the cache only * Cache management: data transfer between cache and primary data * Dirty page: block modified by the client * Flush: dirty blocks written on master copy *Write - delayed write - write to disk if new read needs space (eg: write least recently used block) * Write on close - write when finish with the file *Write-through - write to disk before any further operation Possible places for files in a client-server system - Server disk/main memory - Client disk/main memory
- 11Semantics for sharing - Simultaneous access by multiple clients: Data changes made by a client are observable by other clients - UNIX Semantics - every read sees effects of all previous writes on the file. Clients can share 'current location' pointer - Session semantics - file session: series of file accesses between open and close * Write to an open file is immediately visible for the local client, but invisible to remote clients having the same file open simultaneously * Once a file is closed, changed made to it are visible but only for later starting sessions. Already open sessions are not effected.- Immutable shared file semantics - the content of a file can not be changed after creation. File sharing only in read-only mode
-Transaction-like semantics - using locking methods - 12Caching issues: - Cache size - problem: faster access needs more expensive memory - Cache unit size - too small units - large administrative overhead - Too large units - too few items fit in - Cache location : * local disk - increases reliability and local autonomy * Main memory: fast access, possibly diskless client * address space of the process * kernel address space * cache manager (user level) process
- 13—
- 14—
- 15—
- 16—
Información
- Descripción
- Distributed systems Fichas sobre Lecture 6 - file systems, creado por minal Kotwal el 25/10/2016.
- Creado por
- minal Kotwal
- Visitas
- 0
- Valoración