CSD511 – Distributed Systems
分散式系統
Chapter 8
Distributed File Systems
吳俊興
國立高雄大學 資訊工程學系
Chapter 8 Distributed File Systems
8.1 Introduction
8.2 File service architecture
8.3 Case study: Sun Network File System
8.4 Case study: The Andrew File System
8.5 Enhancements and further developments
8.6 Summary
2
1.1 Introduction
• Sharing of resources is a key goal for distributed
systems
– printers, storages, network bandwidths, memories, …
• Mechanisms for data sharing
– Web servers
– P2P file sharing
– Distributed storage systems
• Distributed file systems
• Distributed object systems
• Goal of distributed file service
– Enable programs to store and access remote files exactly as
they do local ones
3
Figure 8.1 Storage systems and their properties
Sharing Persis- Distributed Consistency Example
tence cache/replicas maintenance
Main memory 1 RAM
File system 1 UNIX file system
Distributed file system Sun NFS
Web Web server
Distributed shared memory Ivy (DSM, Ch. 18)
Remote objects (RMI/ORB) 1 CORBA
Persistent object store 1 CORBA Persistent
Object Service
Peer-to-peer storage system 2 OceanStore (Ch. 10)
Types of consistency:
1: strict one-copy. : slightly weaker guarantees. 2: considerably weaker guarantees.
4
8.1.1 Characteristics of file systems
• File system: responsible for organization, storage,
retrieval, naming, sharing and protection of files
– file: containing data and attributes (Fig 8.3)
– directory: mapping from text names to internal file identifiers
– file operation: system calls in UNIX (Fig. 8.4)
Directory module: relates file names to file IDs
File module: relates file IDs to particular files
Access control module: checks permission for operation requested
File access module: reads or writes file data or attributes
Block module: accesses and allocates disk blocks
Device module: disk I/O and buffering
Figure 8.2 File system modules
5
Figure 8.3 File attribute record structure
File length
Creation timestamp
Read timestamp
Write timestamp
Attribute timestamp
Reference count
Owner
File type
Access control list
6
Figure 8.4 UNIX file system operations
filedes = open(name, mode) Opens an existing file with the given name.
filedes = creat(name, mode) Creates a new file with the given name.
Both operations deliver a file descriptor referencing the open
file. The mode is read, write or both.
status = close(filedes) Closes the open file filedes.
count = read(filedes, buffer, n) Transfers n bytes from the file referenced by filedes to buffer.
count = write(filedes, buffer, n) Transfers n bytes to the file referenced by filedes from buffer.
Both operations deliver the number of bytes actually transferred
and advance the read-write pointer.
pos = lseek(filedes, offset, Moves the read-write pointer to offset (relative or absolute,
whence) depending on whence).
status = unlink(name) Removes the file name from the directory structure. If the file
has no other names, it is deleted.
status = link(name1, name2) Adds a new name (name2) for a file (name1).
status = stat(name, buffer) Gets the file attributes for file name into buffer.
7
8.1.2 Distributed file system requirements
•Transparency:
– access, location, mobility, performance, scaling
•Concurrent file updates
– changes to a file by one client should not interfere with the operation of
other clients simultaneously accessing or changing the same file
– common services: advisory or mandatory file- or record-level locking
•File replication
– A file may be represented by several copies of its contents at different
locations
– advantages: load sharing and fault tolerance
•Hardware and operating system heterogeneity
•Fault tolerance
– The service continue to operate in the face of client and server failures
•Consistency
– An inevitable delay in the propagation of modifications to all sites
•Security
•Efficiency
8
8.1.3 Case studies
• Sun NFS (Network File System)
– introduced in 1985
– the first file service designed as a product
– RFC1813: NFS protocol version 3
– Each computer can act as both a client and a server
• Andrew File System (AFS)
– developed at CMU for use as a campus computing and
information system
– achieved by transferring whole files between server and client
computers and caching them at clients until the server
receives a more up-to-date version
9
8.2 File service architecture
Client computer Server computer
Application Application Directory service
program program
Flat file service
Client module
Figure 8.5 File service architecture (Author’s abstract model)
• Stateless file service architecture
– Flat file service: unique file identifiers (UFID)
– Directory service: map names to UFIDs
– Client module
• integrate/extend flat file and directory services
• provide a common application programming interface (can emulate different file interfaces)
• stores location of flat file and directory services
10
Flat file service interface
• RPC used by client modules, not by user-level programs
• Compared to UNIX
– no open/close
• Create is not idempotent
• at-least-once semantics
• reexecution gets a new file
– specify starting location in Read/Write
• stateless server
Read(FileId, i, n) -> Data If 1 ≤ i ≤ Length(File): Reads a sequence of up to n items
— throws BadPosition from a file starting at item i and returns it in Data.
Write(FileId, i, Data) If 1 ≤ i ≤ Length(File)+1: Writes a sequence of Data to a
— throws BadPosition file, starting at item i, extending the file if necessary.
Create() -> FileId Creates a new file of length 0 and delivers a UFID for it.
Delete(FileId) Removes the file from the file store.
GetAttributes(FileId) -> Attr Returns the file attributes for the file.
SetAttributes(FileId, Attr) Sets the file attributes (only those attributes that are not
shaded in Figure 8.3).
Figure 8.6 Flat file service operations 11
Access control
• UNIX checks access rights when a file is opened
– subsequent checks during read/write are not necessary
• distributed environment
– server has to check
– stateless approaches
1. access check once when UFID is issued
– client gets an encoded "capability" (who can access and how)
– capability is submitted with each subsequent request
2. access check for each request.
– second is more common
12
Directory service operations
• A directory service interface translates text names to file identifiers and
performs a number of other services such as those listed among the sample
commands in figure 8.7. This is the remote procedure call interface to
extend the local directory services to a distributed model.
Lookup(Dir, Name) -> FileId Locates the text name in the directory and returns the
— throwsNotFound relevant UFID. If Name is not in the directory, throws an
exception.
AddName(Dir, Name, FileId) If Name is not in the directory, adds (Name, File) to the
— throwsNameDuplicate directory and updates the file’s attribute record.
If Name is already in the directory: throws an exception.
UnName(Dir, Name) If Name is in the directory: the entry containing Name is
— throwsNotFound removed from the directory.
If Name is not in the directory: throws an exception.
GetNames(Dir, Pattern) -> NameSeq Returns all the text names in the directory that match the
regular expression Pattern.
Figure 8.7 Directory service operations
13
File collections
• Hierarchical file system
– Directories containing other directories and files
– Each file can have more than one name (pathnames)
• how in UNIX, Windows?
• File groups
– a logical collection of files on one server
• a server can have more than one group
• a group can change server
• a file can't change to a new group (copying doesn't count)
• filesystems in unix
• different devices for non-distributed
• different hosts for distributed
32 bits 16 bits
file group identifier IP address date
14
8.3 Case Study: Sun NFS
• Sun NFS
– Industry standard for local networks since the 1980’s
– OS independent (originally unix implementation)
– rpc over udp or tcp
Client computer Server computer
Application Application
program program
UNIX
system calls
UNIX kernel
UNIX kernel Virtual file system Virtual file system
Local Remote
file system
UNIX NFS NFS UNIX
file file
Other
client server
system system
NFS
protocol
Figure 8.8 NFS architecture
15
Schematic view of NFS architecture
Refer to Section 11.9 of the OS Textbook
16
Virtual file system
• Part of unix kernel to support access transparency
• NFS file handles, 3 components:
– i-node (index node)
• structure for finding the file
– filesystem identifier
• different groups of files
– i-node generation number
• i-nodes are reused
• incremented when reused
• VFS
– struct for each file system
– v-node for each open file
• file handle for remote file
• i-node number for local file
17
Access control, client integration, pathname translation
• Access control
– nfs server is stateless, doesn't keep open files for clients
– server check identity each time (uid and gid)
• Client integration
– nfs client emulates Unix file semantics
– in the kernel, not in a library, because:
• access files via system calls
• single client module for multiple user processes
• encryption can be done in the kernel
• Pathname translation
– pathname: /users/students/dc/abc
– server doesn't receive the entire pathname for translation, why?
– client breaks down the pathnames into parts
– iteratively translate each part
– translation is cached
18
Figure 8.9 server operations (simplified)
lookup(dirfh, name) -> fh, attr Returns file handle and attributes for the file name in the directory
dirfh.
create(dirfh, name, attr) -> Creates a new file name in directory dirfh with attributes attr and
newfh, attr returns the new file handle and attributes.
remove(dirfh, name) status Removes file name from directory dirfh.
getattr(fh) -> attr Returns file attributes of file fh. (Similar to the UNIX stat system
call.)
setattr(fh, attr) -> attr Sets the attributes (mode, user id, group id, size, access time and
modify time of a file). Setting the size to 0 truncates the file.
read(fh, offset, count) -> attr, data Returns up to count bytes of data from a file starting at offset.
Also returns the latest attributes of the file.
write(fh, offset, count, data) -> attr Writes count bytes of data to a file starting at offset. Returns the
attributes of the file after the write has taken place.
rename(dirfh, name, todirfh, toname) Changes the name of file name in directory dirfh to toname in
-> status directory to todirfh
.
link(newdirfh, newname, dirfh, name) Creates an entry newname in the directory newdirfh which refers to
-> status file name in the directory dirfh.
19
Figure 8.9 NFS server operations (simplified) cont.
symlink(newdirfh, newname, string) Creates an entry newname in the directory newdirfh of type
-> status symbolic link with the value string. The server does not interpret
the string but makes a symbolic link file to hold it.
readlink(fh) -> string Returns the string that is associated with the symbolic link file
identified by fh.
mkdir(dirfh, name, attr) -> Creates a new directory name with attributes attr and returns the
newfh, attr new file handle and attributes.
rmdir(dirfh, name) -> status Removes the empty directory name from the parent directory dirfh.
Fails if the directory is not empty.
readdir(dirfh, cookie, count) -> Returns up to count bytes of directory entries from the directory
entries dirfh. Each entry contains a file name, a file handle, and an opaque
pointer to the next directory entry, called a cookie. The cookie is
used in subsequent readdir calls to start reading from the following
entry. If the value of cookie is 0, reads from the first entry in the
directory.
statfs(fh) -> fsstats Returns file system information (such as block size, number of
free blocks and so on) for the file system containing a file fh.
20
Mount
• Mount service
– mounting: the process of including a new filesystem
– /etc/exports has filesystems that can be mounted by others
– clients use a modified mount command for remote
filesystems
– communicates with the mount process on the server in a
mount protocol
– hard-mounted
• user process is suspended until request is successful
• when server is not responding
• request is retried until it's satisfied
– soft-mounted
• if server fails, client returns failure after a small # of retries
• user process handles the failure
• Automounter
– what if a user process reference a file on a remote filesystem that is not
mounted
21
– table of mount points (pathname) and servers
Accessible file systems on an NFS client
Server 1 Client Server 2
(root) (root) (root)
export . . . vmunix usr nfs
Remote Remote
people students x staff users
mount mount
big jon bob . . . jim ann jane joe
Note: The file system mounted at /usr/students in the client is actually the sub-tree located at /export/people in Server
the file system mounted at /usr/staff in the client is actually the sub-tree located at /nfs/users in Server 2.
Figure 8.10 Local and remote file systems accessible on an NFS client
22
Caching
• Server caching
– caching file pages, directory/file attributes
– read-ahead: prefetch pages following the most-recently read file pages
– delayed-write: write to disk when the page in memory is needed for other
purposes
– "sync" flushes "dirty" pages to disk every 30 seconds
– two write option
1. write-through: write to disk before replying to the client
2. cache and commit:
– stored in memory cache
– write to disk before replying to a "commit" request from the client
• Client caching
– caches results of read, write, getattr, lookup, readdir
– clients responsibility to poll the server for consistency
23
Client caching: reading
• timestamp-based methods for consistency validation
– Tc: time when the cache entry was last validated
– Tm: time when the block was last modified at the server
• cache entry is valid if:
1. T - Tc < t, where t is the freshness interval
• t is adaptively adjusted:
– files: 3 to 30 seconds depending on freq of updates
– directories: 30 to 60 seconds
2. Tmclient = Tmserver
• cache validation
– need validation for all cache accesses due to no share check
– condition “1" can be determined by the client alone--performed first
– Reducing getattr() to the server [for getting Tmserver]
1. new value of Tmserver is received, apply to all cache entries from the same file
2. piggyback getattr() on file operations
3. adaptive alg for update t
– validation doesn't guarantee the same level of consistency as one-copy
24
Client caching: writing
• dirty: modified page in cache
• flush to disk: file is closed or sync from client
• bio-daemon (block input-output)
– read-ahead: after each read request, request the next file block from the
server as well
– delayed write: after a block is filled, it's sent to the server
– reduce the time to wait for read/write
25
Other optimization
• UDP packet is extended to 9KB to containing entire file block
(8KB for UNIX BSD FFS) and RPC message in a single packet
– Clients and servers of NFSv3 can negotiate sizes larger than 8 KB
• Piggybacked
– File status information cached at clients must be updated at least every 3
seconds for active files
– All operations that refer to files or directories are taken as implicit getattr
requests, and the current attribute values are piggybacked along with the
other results of the operation
26
Security, Performance
• Security
– stateless nfs server
– user's identity in each request
– Kerberos authentication during the mount process, which includes uid and
host address
– server maintain authentication info for the mount
– on each file request, nfs checks the uid and address
– one user per client
• Performance
– overhead/penalty is low
– main problems
• frequent getattr() for cache validation (piggybacking)
• relatively poor performance is write-through is used on the server (delay-
write/commit in current versions)
– write < 5%
– lookup is almost 50% (step by step pathname translation)
27
Summary for NFS
An excellent example of a simple, robust, high-performance distributed
service
• access transparency: same system calls for local or remote files
• location transparency: could have a single name space for all files
(depending on all the clients to agree the same name space)
• mobility transparency: mount table need to be updated on each client (not
transparent)
• scalability: can usually support large loads, add processors, disks, servers...
• file replication: read-only replication, no support for replication of files with
updates
• hardware and OS: many ports
• fault tolerance: stateless and idempotent
• consistency: not quite one-copy for efficiency
• security: added encryption--Kerberos
• efficiency: pretty efficient, wide-spread use
28
8.4 Case Study: the Andrew File System
• Goal: provide transparent access to remote shared files
– like NFS and compatible with NFS
• Differing from NFS
– Primarily attributable to the scalability
• Caching of whole files in client nodes
• Two unusual design characteristics
– Whole file serving: the entire contents of directories and files are
transmitted to client computers
• 64KB chunks in AFS3
– Whole-file caching: clients permanently cache a copy of a file or a chunk
on its local disk
29
Scenario of AFS
• Open a new shared remote file
– A user process issues open() for a file not in the local cache
and then sends a request to the server
– The server returns the requested file
– The copy is stored in the client’s local UNIX file system and
the resulting UNIX file descriptor is returned to the client
• Subsequent read, write and other operations on the
file are applied to the local copy
• When the process in the client issues close()
– if the local copy has been updated, its contents are sent back
to the server
• server updates the contents and the timestamps on the file
• the copy on the client’s local disk is retained
30
Characteristics
• Good for shared files likely to remain valid for long periods
– infrequently updated
– normally accessed by only a single user
Overwhelming majority of file accesses
• Local cache can be allocated a substantial proportion of the
disk space
– should be enough for a working set of files used by one user
• Assumptions about average and maximum file size and
reference locality
– Files are small; most are less than 10KB in size
– Read operations are much more common than writes
– Sequential access is much more common than random access
– Most files are written by only one user. When a file is shared, it is usually
only one user who modified it
– Files are referenced in bursts. A file referenced recently is very probably
referenced soon.
• Maybe good for distributed database applications
31
Venus and Vice: software components in AFS
Workstations Servers
User Venus
program
Vice
UNIX kernel
UNIX kernel
User Venus Network
program
UNIX kernel
Vice
Venus
User
program UNIX kernel
UNIX kernel
Figure 8.11 Distribution of processes in the Andrew File System
32
File name space seen by clients of AFS
Local Shared
/ (root)
tmp bin . . . vmunix cmu
bin
Symbolic
links
A special sub-tree (called /cmu) containing all of the shared files
Figure 8.12 File name space seen by clients of AFS
33
System call interception in AFS
Workstation
User Venus
program
UNIX file Non-local file
system calls operations
UNIX kernel
UNIX file system
Local
disk
Open, close and some other file system calls are intercepted when refer to shared files
Figure 8.13 System call interception in AFS
34
Figure 8.14
Implementation of file system calls in AFS
User process Unix Kernel Venus Net Vice
open(FileName, If FileName refers to a
mode) file in shared file space, Check list of files in local cache. If not
pass the request to present or there is no valid callback promise,
Venus. send a request for the file to the Vice server → Transfer a copy of the file
that is custodian of the volume containing and a callback promise to
Open the local file and the file. the workstation. Log the
return the file callback promise.
descriptor to the Place the copy of the file in the local file
application. system, enter its local name in the local
←
cache list and return the local name to UNIX.
read(FileDescriptor, Perform a normal
Buffer, length) UNIX read operation
on the local copy.
write(FileDescriptor, Perform a normal
Buffer, length) UNIX write operation
on the local copy
close(FileDescriptor) Close the local copy
and notify Venus that If the local copy has been changed, send a
the file has been copy to the Vice serve r that is the custodian Replace the file contents
closed. of the file. → and send a callback to all
other clients holding
callback promises on the
file.
35
Main components of Vice service interface
Fetch(fid) -> attr, data Returns the attributes (status) and, optionally, the contents of file
identified by the fid and records a callback promise on it.
Store(fid, attr, data) Updates the attributes and (optionally) the contents of a specified
file.
Create() -> fid Creates a new file and records a callback promise on it.
Remove(fid) Deletes the specified file.
SetLock(fid, mode) Sets a lock on the specified file or directory. The mode of the
lock may be shared or exclusive. Locks that are not removed
expire after 30 minutes.
ReleaseLock(fid) Unlocks the specified file or directory.
RemoveCallback(fid) Informs server that a Venus process has flushed a file from its
cache.
BreakCallback(fid) This call is made by a Vice server to a Venus process. It cancels
the callback promise on the relevant file.
Figure 8.15
36
8.5 Enhancements and further developments
NFS enhancements
WebNFS - NFS server implements a web-like service on a well-known
port. Requests use a 'public file handle' and a pathname-capable
variant of lookup(). Enables applications to access NFS servers
directly, e.g. to read a portion of a large file.
One-copy update semantics (Spritely NFS, NQNFS) - Include an open()
operation and maintain tables of open files at servers, which are used
to prevent multiple writers and to generate callbacks to clients
notifying them of updates. Performance was improved by reduction in
gettattr() traffic.
Improvements in disk storage organisation
RAID - improves performance and reliability by striping data redundantly
across several disk drives
Log-structured file storage - updated pages are stored contiguously in
memory and committed to disk in large contiguous blocks (~ 1 Mbyte).
File maps are modified whenever an update occurs. Garbage
collection to recover disk space.
37
New design approaches
• Distribute file data across several servers
– Exploits high-speed networks (ATM, Gigabit Ethernet)
– Layered approach, lowest level is like a 'distributed virtual disk'
– Achieves scalability even for a single heavily-used file
• 'Serverless' architecture
– Exploits processing and disk resources in all available network nodes
• Service is distributed at the level of individual files
– Examples:
• xFS (section 8.5): Experimental implementation demonstrated a substantial
performance gain over NFS and AFS
• Frangipani (section 8.5): Performance similar to local UNIX file access
• Tiger Video File System (see Chapter 15)
• P2P systems: Napster, OceanStore (UCB), Farsite (MSR), Publius (AT&T
research) - see web for documentation on these very recent systems
• Replicated read-write files
– High availability
– Disconnected working
• re-integration after disconnection is a major problem if conflicting updates have
occurred 38
