Transactions and Reliability Sarah Diesburg Operating Systems CS
Transactions and Reliability Sarah Diesburg Operating Systems CS 3430
Motivation l File systems have lots of metadata: ¡Free blocks, directories, file headers, indirect blocks l Metadata is heavily cached for performance
Problem l System crashes l OS needs to ensure that the file system does not reach an inconsistent state l Example: move a file between directories ¡Remove a file from the old directory ¡Add a file to the new directory l What happens when a crash occurs in the middle?
UNIX File System (Ad Hoc Failure. Recovery) l Metadata handling: ¡Uses a synchronous write-through caching policy l. A call to update metadata does not return until the changes are propagated to disk ¡Updates are ordered ¡When crashes occur, run fsck to repair inprogress operations
Some Examples of Metadata Handling l Undo effects not yet visible to users ¡If a new file is created, but not yet added to the directory l. Delete the file l Continue effects that are visible to users ¡If file blocks are already allocated, but not recorded in the bitmap l. Update the bitmap
UFS User Data Handling l Uses a write-back policy ¡Modified blocks are written to disk at 30 -second intervals l. Unless a user issues the sync system call ¡Data updates are not ordered ¡In many cases, consistent metadata is good enough
Example: Vi l Vi saves changes by doing the following 1. Writes the new version in a temp file l. Now we have old_file and new_temp file 2. Moves the old version to a different temp file l. Now we have new_temp and old_temp 3. Moves the new version into the real file l. Now we have new_file and old_temp 4. Removes the old version l. Now we have new_file
Example: Vi l When crashes occur ¡Looks for the leftover files ¡Moves forward or backward depending on the integrity of files
Transaction Approach l A transaction groups operations as a unit, with the following characteristics: ¡Atomic: all operations either happen or they do not (no partial operations) ¡Serializable: transactions appear to happen one after the other ¡Durable: once a transaction happens, it is recoverable and can survive crashes
More on Transactions l A transaction is not done until it is committed l Once committed, a transaction is durable l If a transaction fails to complete, it must rollback as if it did not happen at all l Critical sections are atomic and serializable, but not durable
Transaction Implementation (One Thread) l Example: money transfer Begin transaction x = x – 1; y = y + 1; Commit
Transaction Implementation (One Thread) l Common implementations involve the use of a log, a journal that is never erased l A file system uses a write-ahead log to track all transactions
Transaction Implementation (One Thread) l Once accounts of x and y are on a log, the log is committed to disk in a single write l Actual changes to those accounts are done later
Transaction Illustrated x = 1; y = 1;
Transaction Illustrated x = 0; y = 2; x = 1; y = 1;
Transaction Illustrated x = 0; y = 2; begin transaction old x: 1 new x: 0 old y: 1 new y: 2 commit x = 1; y = 1; Commit the log to disk before updating the actual values on disk
Transaction Steps l Mark the beginning of the transaction l Log the changes in account x l Log the changes in account y l Commit l Modify account x on disk l Modify account y on disk
Scenarios of Crashes l If a crash occurs after the commit ¡Replays the log to update accounts l If a crash occurs before or during the commit ¡Rolls back and discard the transaction
Two-Phase Locking (Multiple Threads) l Logging alone not enough to prevent multiple transactions from trashing one another (not serializable) l Solution: two-phase locking 1. Acquire all locks 2. Perform updates and release all locks l Thread A cannot see thread B’s changes until thread A commits and releases locks
Transactions in File Systems l Almost all file systems built since 1985 use write-ahead logging ¡NTFS, HFS+, ext 3, ext 4, … + Eliminates running fsck after a crash + Write-ahead logging provides reliability - All modifications need to be written twice
Log-Structured File System (LFS) l If logging is so great, why don’t we treat everything as log entries? l Log-structured file system ¡Everything is a log entry (file headers, directories, data blocks) ¡Write the log only once l. Use version stamps to distinguish between old and new entries
More on LFS l New log entries are always appended to the end of the existing log ¡All writes are sequential ¡Seeks only occurs during reads l. Not so bad due to temporal locality and caching l Problem: ¡Need to create more contiguous space all the time
RAID and Reliability l So far, we assume that we have a single disk l What if we have multiple disks? ¡ The chance of a single-disk failure increases l RAID: redundant array of independent disks ¡ Standard way of organizing disks and classifying the reliability of multi-disk systems ¡ General methods: data duplication, parity, and errorcorrecting codes (ECC)
RAID 0 l No redundancy l Uses block-level striping across disks ¡i. e. , 1 st block stored on disk 1, 2 nd block stored on disk 2 l Failure causes data loss
Non-Redundant Disk Array Diagram (RAID Level 0) open(foo) read(bar) write(zoo) File System
Mirrored Disks (RAID Level 1) l Each disk has a second disk that mirrors its contents ¡Writes go to both disks + Reliability is doubled + Read access faster - Write access slower - Expensive and inefficient
Mirrored Disk Diagram (RAID Level 1) open(foo) read(bar) write(zoo) File System
Memory-Style ECC Diagram (RAID Level 2) open(foo) read(bar) write(zoo) File System
Byte-Interleaved Parity (RAID Level 3) l Uses bye-level striping across disks ¡i. e. , 1 st byte stored on disk 1, 2 nd byte stored on disk 2 l One disk in the array stores parity for the other disks ¡Parity can be used to recover bits on a lost disk ¡No detection bits needed, relies on disk controller to detect errors + More efficient than Levels 1 and 2 - Parity disk doesn’t add bandwidth
Parity Method l Disk 1: l Disk 2: l Disk 3: l Parity: 1001 0101 1000 0100 = 1001 xor 0101 xor 1000 l To recover disk 2 ¡Disk 2: 0101 = 1001 xor 1000 xor 0100
Byte-Interleaved RAID Diagram (Level 3) open(foo) read(bar) write(zoo) File System
Block-Interleaved Parity (RAID Level 4) l Like byte-interleaved, but data is interleaved in blocks + More efficient data access than level 3 - Parity disk can be a bottleneck - Small writes require 4 I/Os ¡Read the old block ¡Read the old parity ¡Write the new block ¡Write the new parity
Block-Interleaved Parity Diagram (RAID Level 4) open(foo) read(bar) write(zoo) File System
Block-Interleaved Distributed-Parity (RAID Level 5) l Sort of the most general level of RAID l Spreads the parity out over all disks + No parity disk bottleneck + All disks contribute read bandwidth – Requires 4 I/Os for small writes
Block-Interleaved Distributed-Parity Diagram (RAID Level 5) open(foo) read(bar) write(zoo) File System
Other RAIDs l RAID 6 ¡Extends RAID 5 by adding another parity block ¡Uses block-level striping with two parity blocks across all disks ¡ 4 drive minimum, can survive two drive failures l RAID 10 (RAID 1 + RAID 0) ¡Combines disk mirroring and striping l RAID 50 (RAID 5 + RAID 0) ¡Combines distributed parity with striping
- Slides: 36