Outline for Today’s Lecture

Administrative:– Welcome back!– Programming assignment 2 due date

extended (as requested)

Objective: – File caching– NTFS– Distributed File Systems

File Buffer Cache

• Avoid the disk for as many file operations as possible.

• Cache acts as a filter for the requests seen by the disk

• Reads well-served. Reuse!• Delayed writeback will avoid

going to disk at all for temp files.

Memory

Filecache

Proc

Why Are File Caches Effective?

1. Locality of reference: storage accesses come in clumps.spatial locality: If a process accesses data in block B, it is likely

to reference other nearby data soon.(e.g., the remainder of block B)

example: reading or writing a file one byte at a time

temporal locality: Recently accessed data is likely to be used again.

2. Read-ahead: if we can predict what blocks will be needed soon, we can prefetch them into the cache.most files are accessed sequentially

File Buffer Cache

Memory

Filecache

Proc

V.M.

What do wewant to cache?

• inodes• dir nodes• whole files• disk blocks

The answer will determine• where in the request path the cache sits

Access Patterns Along the Way

File cache

Proc F.S.

open(/foo/bar/file);read(fd,buf,sizeof(buf));read(fd,buf,sizeof(buf));read(fd,buf,sizeof(buf));close(fd);

read(rootdir);read(inode);read(foo);read(inode);read(bar);read(inode);read(fd,buf,sizeof(buf));read(fd,buf,sizeof(buf));read(fd,buf,sizeof(buf));

read(rootdir);read(inode);read(foo);read(inode);read(bar);read(inode);read(filedatablock);

File Access Patterns

• What do users seem to want from the file abstraction?

• What do these usage patterns mean for file structure and implementation decisions?– What operations should be optimized 1st?– How should files be structured?– Is there temporal locality in file usage?– How long do files really live?

Know your Workload!

• File usage patterns should influence design decisions. Do things differently depending:– How large are most files? How long-lived?

Read vs. write activity. Shared often?– Different levels “see” a different workload.

• Feedback loop

Usage patterns observed today

File Systemdesign and impl

What to Cache? Locality in File Access Patterns

(UNIX Workloads)

• Most files are small (often fitting into one disk block) although most bytes are transferred from longer files.

• Accesses tend to be sequential and 100%– Spatial locality– What happens when we cache a huge file?

• Most opens are for read mode, most bytes transferred are by read operations

• There is significant reuse (re-opens) most opens go to files repeatedly opened & quickly. Directory nodes and executables also exhibit good temporal locality.– Looks good for caching!

• Use of temp files is significant part of file system activity in UNIX very limited reuse, short lifetimes (less than a minute).

• Long absolute pathnames are common in file opens – Name resolution can dominate performance – why?

What to Cache? Locality in File Access Patterns

(continued)

What to do about long paths?

• Make long lookups cheaper cluster inodes and data on disk to make each component resolution step somewhat cheaper– Immediate files meta-data and first block of data

co-located

• Collapse prefixes of paths hash table– Prefix table

• “Cache it” in this case, directory info

Issues for Implementing an I/O Cache Structure

Goal: maintain K slots in memory as a cache over a collection of m items on secondary storage (K << m).

1. What happens on the first access to each item?Fetch it into some slot of the cache, use it, and leave it there to

speed up access if it is needed again later.

2. How to determine if an item is resident in the cache?Maintain a directory of items in the cache: a hash table.Hash on a unique identifier (tag) for the item (fully associative).

3. How to find a slot for an item fetched into the cache?Choose an unused slot, or select an item to replace according to

some policy, and evict it from the cache, freeing its slot.

File Block Buffer Cache

HASH(vnode, logical block) Buffers with valid data are retained in memory in a buffer cache or file cache.

Each item in the cache is a buffer header pointing at a buffer .

Blocks from different files may be intermingled in the hash chains.

System data structures hold pointers to buffers only when I/O is pending or imminent.

- busy bit instead of refcount

- most buffers are “free”

free/inactivelist head

free/inactive list tail

Most systems use a pool of buffers in kernel memory as a staging area for memory<->disk transfers.

Handling Updates in the File Cache

1. Blocks may be modified in memory once they have been brought into the cache.

Modified blocks are dirty and must (eventually) be written back.Write-back, write-through (104?)

2. Once a block is modified in memory, the write back to disk may not be immediate (synchronous).

Delayed writes absorb many small updates with one disk write.How long should the system hold dirty data in memory?

Asynchronous writes allow overlapping of computation and disk update activity (write-behind).Do the write call for block n+1 while transfer of block n is in progress.

Thus file caches also can improve performance for writes.

3. Knowing data gets to disk Force it but you can’t trust to a “write” syscall - fsync

Mechanism for Cache Eviction/Replacement

• Typical approach: maintain an ordered free/inactive list of slots that are candidates for reuse.– Busy items in active use are not on the list.

• E.g., some in-memory data structure holds a pointer to the item.

• E.g., an I/O operation is in progress on the item.

– The best candidates are slots that do not contain valid items.• Initially all slots are free, and they may become free again as items

are destroyed (e.g., as files are removed).

– Other slots are listed in order of value of the items they contain.• These slots contain items that are valid but inactive: they are held

in memory only in the hope that they will be accessed again later.

Replacement Policy• The effectiveness of a cache is

determined largely by the policy for ordering slots/items on the free/inactive list. Defines the replacement policy

• A typical cache replacement policy is LRU– Assume hot items used recently are likely

to be used again.– Move the item to the tail of the free list on

every release.– The item at the front of the list is the

coldest inactive item.

• Other alternatives:– FIFO: replace the oldest item.– MRU/LIFO: replace the most recently used

item.

HASH(vnode, logical block)

free/inactivelist head

free/inactive list tail

Viewing Memory as a Unified I/O Cache

A key role of the I/O system is to manage the page/block cache for performance and reliability.

tracking cache contents and managing page/block sharing

choreographing movement to/from external storage

balancing competing uses of memory

Modern systems attempt to balance memory usage between the VM system and the file cache.

Grow the file cache for file-intensive workloads.

Grow the VM page cache for memory-intensive workloads.

Support a consistent view of files across different style of access.

unified buffer cache

Synchronization Problems for a Cache

1. What if two processes try to get the same block concurrently, and the block is not resident?

2. What if a process requests to write block A while a put is already in progress on block A?

3. What if a get must replace a dirty block A in order to allocate a buffer to fetch block B?

This will happen if the block/buffer at the head of the free list is dirty.

What if another process requests to get A during the put?

4. How to handle read/write requests on shared files atomically?Unix guarantees that a read will not return the partial result of a

concurrent write, and that concurrent writes do not interleave.

Linux Page Cache• Page Cache is the disk cache for all page-

based I/O – subsumes file buffer cache.– All page I/O flows through page cache

• pdflush daemons – writeback to disk any dirty pages/buffers.– When free memory falls below threshold, wakeup

daemon to reclaim free memory• Specified number written back• Free memory above threshold

– Periodically, to prevent old data not getting written back, wakeup on timer expiration

• Writes all pages older than specified limit.

Layout

Layout on Disk

• Can address both seek and rotational latency• Cluster related things together

(e.g. an inode and its data, inodes in same directory (ls command), data blocks of multi-block file, files in same directory)

• Sub-block allocation to reduce fragmentation for small files

• Log-Structured File Systems

File Structure Implementation:

Mapping File Block• Contiguous

– 1 block pointer, causes fragmentation, growth is a problem.

• Linked– each block points to next block, directory points to

first, OK for sequential access

• Indexed– index structure required, better for random access

into file.

UNIX Inodes

FileAttributes

Blo

ck A

ddr

...

......

...

...

... ...

Data Block Addr

1

1

1

2

2

2

2

3 3 3 3

Data blocks

Decoupling meta-datafrom directory entries

File Allocation Table (FAT)

Lecture.ppt

Pic.jpg

Notes.txt

eof

eof

eof

The Problem of Disk Layout• The level of indirection in the file block maps

allows flexibility in file layout.• “File system design is 99% block allocation.” [McVoy]

• Competing goals for block allocation:– allocation cost– bandwidth for high-volume transfers– efficient directory operations

• Goal: reduce disk arm movement and seek overhead.

• metric of merit: bandwidth utilization

FFS and LFSTwo different approaches to block allocation:

– Cylinder groups in the Fast File System (FFS) [McKusick81]

• clustering enhancements [McVoy91], and improved cluster allocation [McKusick: Smith/Seltzer96]

• FFS can also be extended with metadata logging [e.g., Episode]

– Log-Structured File System (LFS)• proposed in [Douglis/Ousterhout90]• implemented/studied in [Rosenblum91]• BSD port, sort of maybe: [Seltzer93]• extended with self-tuning methods [Neefe/Anderson97]

– Other approach: extent-based file systems

Log-Structured File Systems• Assumption: Cache is effectively filtering out reads so we

should optimize for writes• Basic Idea: manage disk as an append-only log (subsequent

writes involve minimal head movement)• Data and meta-data (mixed) accumulated in large segments

and written contiguously• Reads work as in UNIX - once inode is found, data blocks

located via index.• Cleaning an issue - to produce contiguous free space,

correcting fragmentation developing over time.• Claim: LFS can use 70% of disk bandwidth for writing while

Unix FFS can use only 5-10% typically because of seeks.

LFS logs

In LFS, all block and metadata allocation is log-based.– LFS views the disk as “one big log” (logically).– All writes are clustered and sequential/contiguous.

• Intermingles metadata and blocks from different files.– Data is laid out on disk in the order it is written.– No-overwrite allocation policy: if an old block or inode is

modified, write it to a new location at the tail of the log.– LFS uses (mostly) the same metadata structures as FFS;

only the allocation scheme is different.• Cylinder group structures and free block maps are

eliminated.• Inodes are found by indirecting through a new map

LFS Data Structures on Disk

• Inode – in log, same as FFS• Inode map – in log, locates position of inode, version,

time of last access• Segment summary – in log, identifies contents of

segment (file#, offset for each block in segment)• Segment usage table – in log, counts live bytes in

segment and last write time• Checkpoint region – fixed location on disk, locates

blocks of inode map, identifies last checkpoint in log.• Directory change log – in log, records directory

operations to maintain consistency of ref counts in inodes

Structure of the Log

clean

Checkpoint region

= inode map block

= inode

D1

= directory node

File 2File 1

block 2

= data block

File 1block1

= segment summary, usage, dirlog

Writing the Log in LFS1. LFS “saves up” dirty blocks and dirty inodes until it

has a full segment (e.g., 1 MB).– Dirty inodes are grouped into block-sized clumps.– Dirty blocks are sorted by (file, logical block number).– Each log segment includes summary info and a

checksum.

2. LFS writes each log segment in a single burst, with at most one seek.– Find a free segment “slot” on the disk, and write it.– Store a back pointer to the previous segment.

• Logically the log is sequential, but physically it consists of a chain of segments, each large enough to amortize seek overhead.

Growth of the Log

write (file1, block1)creat (D1/file3)write (file3, block1)

clean

Checkpoint region

D1File 2File 1

block 2File 1block1

File 1block1

File 3 D1

Death in the Log

write (file1, block1)creat (D1/file3)write (file3, block1)

clean

Checkpoint region

D1File 2File 1

block 2File 1block1

File 1block1

File 3 D1

Writing the Log: the Rest of the Story

1. LFS cannot always delay writes long enough to accumulate a full segment; sometimes it must push a partial segment.– fsync, update daemon, NFS server, etc.– Directory operations are synchronous in FFS, and some must

be in LFS as well to preserve failure semantics and ordering.

2. LFS allocation and write policies affect the buffer cache, which is supposed to be filesystem-independent.– Pin (lock) dirty blocks until the segment is written; dirty blocks

cannot be recycled off the free chain as before.– Endow *indirect blocks with permanent logical block numbers

suitable for hashing in the buffer cache.

Cleaning in LFSWhat does LFS do when the disk fills up?1. As the log is written, blocks and inodes written earlier in

time are superseded (“killed”) by versions written later.– files are overwritten or modified; inodes are updated– when files are removed, blocks and inodes are deallocated

2. A cleaner daemon compacts remaining live data to free up large hunks of free space suitable for writing segments.– look for segments with little remaining live data

• benefit/cost analysis to choose segments

– write remaining live data to the log tail– can consume a significant share of bandwidth, and there are

lots of cost/benefit heuristics involved.

Cleaning the Log

Checkpoint region

D1File 2File 1

block 2File 1block1

File 1block1 File 3 D1

clean

Cleaning the Log

Checkpoint region

D1

File 2

File 1 block 2

File 1block1

File 1block1 File 3 D1

File 1 block 2

File 2

Cleaning the Log

Checkpoint region

File 2

File 1block1 File 3 D1

File 1 block 2

clean

Cleaning Issues

• Must be able to identify which blocks are live• Must be able to identify the file to which each

block belongs in order to update inode to new location

• Segment Summary block contains this info– File contents associated with uid (version # and

inode #)– Inode entries contain version # (incr. on truncate)– Compare to see if inode points to block under

consideration

Policies• When cleaner cleans – threshold based• How much – 10s at a time until threshold reached• Which segments

– Most fragmented segment is not best choice.– Value of free space in segment depends on stability of

live data (approx. age)– Cost / benefit analysis Benefit = free space available (1-u) * age of youngest block

Cost = cost to read segment + cost to move live data

– Segment usage table supports this

• How to group live blocks

Recovering Disk Contents

• Checkpoints – define consistent states– Position in log where all data structures are consistent– Checkpoint region (fixed location) – contains the addresses of all

blocks of inode map and segment usage table, ptr to last segment written

• Actually 2 that alternate in case a crash occurs while writing checkpoint region data

• Roll-forward – to recover beyond last checkpoint– Uses Segment summary blocks at end of log – if we find new

inodes, update inode map found from checkpoint– Adjust utilizations in segment usage table– Restore consistency in ref counts within inodes and directory

entries pointing to those inodes using Directory operation log (like an intentions list)

Recovery of the Log

Checkpoint region

D1File 2File 1

block 2File 1block1

File 1block1

File 3 D1

Written since checkpoint

Recovery in Unix

fsck

• Traverses the directory structure checking ref counts of inodes

• Traverses inodes and freelist to check block usage of all disk blocks

Evaluation of LFS vs. FFS

1. How effective is FFS clustering in “sequentializing” disk writes? Do we need LFS once we have clustering?– How big do files have to be before FFS matches LFS?– How effective is clustering for bursts of creates/deletes?– What is the impact of FFS tuning parameters?

2. What is the impact of file system age and high disk space utilization?– LFS pays a higher cleaning overhead.– In FFS fragmentation compromises clustering effectiveness.

3. What about workloads with frequent overwrites and random access patterns (e.g., transaction processing)?

Which kinds of locality on disk are better?

Benchmarks and Conclusions

1. For bulk creates/deletes of small files, LFS is an order of magnitude better than FFS, which is disk-limited.

• LFS gets about 70% of disk bandwidth for creates.

2. For bulk creates of large files, both FFS and LFS are disk-limited.

3. FFS and LFS are roughly equivalent for reads of files in create order, but FFS spends more seek time on large files.

4. For file overwrites in create order, FFS wins for large files.

The Cleaner Controversy

Seltzer measured TP performance using a TPC-B benchmark (banking application) with a separate log disk.

1. TPC-B is dominated by random reads/writes of account file.

2. LFS wins if there is no cleaner, because it can sequentialize the random writes.

• Journaling log avoids the need for synchronous writes.

3. Since the data dies quickly in this application, LFS cleaner is kept busy, leading to high overhead.

4. Claim: cleaner consumes 34% of disk bandwidth at 48% space utilization, removing any advantage of LFS.

NTFS

NTFS

• API functions for file I/O in Windows 2000• Second column gives nearest UNIX equivalent

File System API Example

A program fragment for copying a file using the Windows 2000 API functions

Directory System Calls

• API functions for directory management in Windows 2000• Second column gives nearest UNIX equivalent, when one exists

File System Structure

The NTFS master file table

File System Structure

The attributes used in MFT records

File System Structure

An MFT record for a three-run, nine-block file

File System Structure

A file that requires three MFT records to store its runs

File System Structure

The MFT record for a small directory.

File Name Lookup

Steps in looking up the file C:\maria\web.htm

Object Mgrname space

File Compression

(a) An example of a 48-block file being compressed to 32 blocks(b) The MTF record for the file after compression

File Encryption

Operation of the encrypting file system

K retrieved

user's public key

Comparisons

FFS LFS NTFS

Data blocks

Clustering,

Cylinder grouping

Log: contiguous by temporal ordering

Runs of contiguous blocks possible, immed.

Directories Directory nodes, cylinder grouping

Directory nodes, in log

They are MFT entries

Block indices

Inodes, specified loc in cylinder group

Inodes in log In MFT entries for files

Distributed File Systems

Distributed File Systems

• Naming– Location

transparency/ independence

• Caching– Consistency

• Replication– Availability and

updates

server

network

server

client

client

client

Naming

• \\His\d\pictures\castle.jpg– Not location transparent - both

machine and drive embedded in name.

• NFS mounting– Remote directory mounted

over local directory in local naming hierarching.

– /usr/m_pt/A– No global view

Her local directory tree

usr

m_pt

His localdir tree

for_export

A B

usr

m_pt

A B

Her local tree after mount A B

usr

m_pt

His after mount on B

Global Name Space

Example: Andrew File System

/

afs

tmp bin lib

local files

shared files -looks identical toall clients

Hints

• A valuable distributed systems design technique that can be illustrated in naming.

• Definition: information that is not guaranteed to be correct. If it is, it can improve performance. If not, things will still work OK. Must be able to validate information.

• Example: Sprite prefix tables

Prefix Tables

m_pt1

usr

m_pt2

A

/

/A/m_pt1/usr/m_pt2 pink

/A/m_pt1 blue

/A/m_pt1/usr/B pink

B

/A/m_pt1/usr/m_pt2/stuff.below

VFS: the Filesystem Switch

syscall layer (file, uio, etc.)

user space

Virtual File System (VFS)networkprotocol

stack(TCP/IP) NFS FFS LFS etc.*FS etc.

device drivers

Sun Microsystems introduced the virtual file system framework in 1985 to accommodate the Network File System cleanly.

• VFS allows diverse specific file systems to coexist in a file tree, isolating all FS-dependencies in pluggable filesystem modules.

VFS was an internal kernel restructuringwith no effect on the syscall interface.

Incorporates object-oriented concepts:a generic procedural interface withmultiple implementations.

Other abstract interfaces in the kernel: device drivers,file objects, executable files, memory objects.

VnodesIn the VFS framework, every file or directory in active

use is represented by a vnode object in kernel memory.

syscall layer

NFS UFS

free vnodes

Active vnodes are reference-counted by the structures thathold pointers to them, e.g.,the system open file table.

Each vnode has a standardfile attributes struct.

Vnode operations aremacros that vector tofilesystem-specificprocedures.

Generic vnode points atfilesystem-specific struct(e.g., inode, rnode), seenonly by the filesystem.

Each specific file system maintains a hash of its resident vnodes.

Vnode Operations and Attributes

directories onlyvop_lookup (OUT vpp, name)vop_create (OUT vpp, name, vattr)vop_remove (vp, name)vop_link (vp, name)vop_rename (vp, name, tdvp, tvp, name)vop_mkdir (OUT vpp, name, vattr)vop_rmdir (vp, name)vop_readdir (uio, cookie)vop_symlink (OUT vpp, name, vattr, contents)vop_readlink (uio)

files onlyvop_getpages (page**, count, offset)vop_putpages (page**, count, sync, offset)vop_fsync ()

vnode/file attributes (vattr or fattr)type (VREG, VDIR, VLNK, etc.)mode (9+ bits of permissions)nlink (hard link count)owner user IDowner group IDfilesystem IDunique file IDfile size (bytes and blocks)access timemodify timegeneration number

generic operationsvop_getattr (vattr)vop_setattr (vattr)vhold()vholdrele()

Pathname Traversal• When a pathname is passed as an argument to a

system call, the syscall layer must “convert it to a vnode”.

• Pathname traversal is a sequence of vop_lookup calls to descend the tree to the named file or directory.

open(“/tmp/zot”)vp = get vnode for / (rootdir)vp->vop_lookup(&cvp, “tmp”);vp = cvp;vp->vop_lookup(&cvp, “zot”);

Issues:1. crossing mount points2. obtaining root vnode (or current dir)3. finding resident vnodes in memory4. caching name->vnode translations5. symbolic (soft) links6. disk implementation of directories7. locking/referencing to handle races with name create and delete operations

Example:Network File System (NFS)

syscall layer

UFS

NFSserver

VFS

VFS

NFSclient

UFS

syscall layer

client

user programs

network

server

NFS Protocol

NFS is a network protocol layered above TCP/IP.– Original implementations (and most today) use UDP

datagram transport for low overhead.• Maximum IP datagram size was increased to match FS

block size, to allow send/receive of entire file blocks.• Some newer implementations use TCP as a transport.

NFS protocol is a set of message formats and types.

• Client issues a request message for a service operation.• Server performs requested operation and returns a reply

message with status and (perhaps) requested data.

File HandlesQuestion: how does the client tell the server which

file or directory the operation applies to?– Similarly, how does the server return the result of a lookup?

• More generally, how to pass a pointer or an object reference as an argument/result of an RPC call?

In NFS, the reference is a file handle or fhandle, a 32-byte token/ticket whose value is determined by the server.– Includes all information needed to identify the

file/object on the server, and get a pointer to it quickly.

volume ID inode # generation #

NFS: From Concept to Implementation

Now that we understand the basics, how do we make it work in a real system?– How do we make it fast?

• Answer: caching, read-ahead, and write-behind.

– How do we make it reliable? What if a message is dropped? What if the server crashes?

• Answer: client retransmits request until it receives a response.

– How do we preserve file system semantics in the presence of failures and/or sharing by multiple clients?

• Answer: well, we don’t, at least not completely.

– What about security and access control?

Distributed File Systems

• Naming– Location

transparency/ independence

• Caching– Consistency

• Replication– Availability and

updates

server

network

server

client

client

client

Caching was “The Answer”

• Avoid the disk for as many file operations as possible.

• Cache acts as a filter for the requests seen by the disk reads served best.

• Delayed writeback will avoid going to disk at all for temp files.

Memory

Filecache

Proc

Caching in Distributed F.S.

• Location of cache on client - disk or memory

• Update policy– write through– delayed writeback– write-on-close

• Consistency– Client does validity check,

contacting server– Server call-backs

server

network

server

client

client

client

File Cache Consistency

Caching is a key technique in distributed systems.The cache consistency problem: cached data may become

stale if cached data is updated elsewhere in the network.

Solutions:Timestamp invalidation (NFS).

Timestamp each cache entry, and periodically query the server: “has this file changed since time t?”; invalidate cache if stale.

Callback invalidation (AFS).Request notification (callback) from the server if the file

changes; invalidate cache on callback.

Leases (NQ-NFS) [Gray&Cheriton89]

89

Sun NFS Cache Consistency• Server is stateless• Requests are self-

contained.• Blocks are transferred and

cached in memory.• Timestamp of last known

mod kept with cached file, compared with “true” timestamp at server on Open. (Good for an interval)

• Updates delayed but flushed before Close ends.

server

network

server

client

client

client

ti

tj

openti== tj ?

write/close

90

Cache Consistency for the Web

• Time-to-Live (TTL) fields - HTTP “expires” header

• Client polling -HTTP “if-modified-since” request headers– polling frequency?

possibly adaptive (e.g. based on age of object and assumed stability)

network

Webserver

proxycache

lan

clientclient

91

AFS Cache Consistency• Server keeps state of all

clients holding copies (copy set)

• Callbacks when cached data are about to become stale

• Large units (whole files or 64K portions)

• Updates propagated upon close

• Cache on local disk memory

server

network

server

c0

c1

c2

{c0, c1}

close

callback

• If client crashes, revalidation on recovery (lost callback possibility)

NQ-NFS Leases

In NQ-NFS, a client obtains a lease on the file that permits the client’s desired read/write activity.

“A lease is a ticket permitting an activity; the lease is valid until some expiration time.”

– A read-caching lease allows the client to cache clean data.Guarantee: no other client is modifying the file.

– A write-caching lease allows the client to buffer modified data for the file.

Guarantee: no other client has the file cached.Leases may be revoked by the server if another client requests a

conflicting operation (server sends eviction notice).Since leases expire, losing “state” of leases at server is OK.

Coda – Using Caching to Handle Disconnected Access • Single location-transparent UNIX FS.• Scalability - coarse granularity

(whole-file caching, volume management)• First class (server) replication and

client caching (second class replication)• Optimistic replication & consistency

maintenance. Designed for disconnected operation for

mobile computing clients

Explicit First-class Replication

• File name maps to set of replicas, one of which will be used to satisfy request– Goal: availability

• Update strategy– Atomic updates - all or none– Primary copy approach– Voting schemes– Optimistic, then detection of conflicts

Optimistic vs. Pessimistic

• High availability Conflicting updates are the potential problem - requiring detection and resolution.

• Avoids conflicts by holding of shared or exclusive locks.

• How to arrange when disconnection is involuntary?

• Leases [Gray, SOSP89]

puts a time-bound on locks but what about expiration?

Client-cache State Transitions

emulation reintegration

hoarding

physicalreconnection

logical reconnection

disconnection

Prefetching

• To avoid the access latency of moving the data in for that first cache miss.

• Prediction! “Guessing” what data will be needed in the future.– It’s not for free:

Consequences of guessing wrongOverhead

Hoarding - Prefetching for Disconnected Information

Access• Caching for availability (not just latency)• Cache misses, when operating disconnected,

have no redeeming value. (Unlike in connected mode, they can’t be used as the triggering mechanism for filling the cache.)

• How to preload the cache for subsequent disconnection? Planned or unplanned.

• What does it mean for replacement?

Hoard Database

• Per-workstation, per-user set of pathnames with priority

• User can explicitly tailor HDB using scripts called hoard profiles

• Delimited observations of reference behavior (snapshot spying with bookends)

Coda Hoarding State

• Balancing act - caching for 2 purposes at once:– performance of current accesses, – availability of future disconnected access.

• Prioritized algorithm - Priority of object for retention in cache is

f(hoard priority, recent usage). • Hoard walking (periodically or on request)

maintains equilibrium - no uncached object has higher priority than any of cached objects

The Hoard Walk

• Hoard walk - phase 1 - reevaluate name bindings (e.g., any new children created by other clients?)

• Hoard walk - phase 2 - recalculate priorities in cache and in HDB, evict and fetch to restore equilibrium

Hierarchical Cache Mgt

• Ancestors of a cached object must be cached in order to resolve pathname.

• Directories with cached children are assigned infinite priority

Callbacks During Hoarding

• Traditional callbacks - invalidate object and refetch on demand

• With threat of disconnection– Purge files and refetch on demand or

hoard walk– Directories - mark as stale and fix on

reference or hoard walk, available until then just in case.

Emulation State

• Pseudo-server, subject to validation upon reconnection

• Cache management by priority– modified objects assigned infinite priority– freeing up disk space - compression, replacement

to floppy, backout updates

• Replay log also occupies non-volatile storage (RVM - recoverable virtual memory)

Client-cache State Transitions with Weak

Connectivity

emulation write disconnected

hoarding

physicalreconnection

strong connection

disconnection weakconnection

disconnection

Cache Misses with Weak Connectivity

• At least now it’s possible to service misses but $$$ and it’s a foreground activity (noticable impact). Maybe not

• User patience threshold - estimated service time compared with what is acceptable

• Defer misses by adding to HDB and letting hoard walk deal with it

• User interaction during hoard walk.