AdvFS/Advanced File System Ccncepts
0 likes•51 views
The document discusses advanced file system concepts, specifically the Architecture of the Advanced File System (advfs), covering terms such as file domains, filesets, and volumes, alongside features like extent-based storage, logging for crash recovery, and cloning capabilities. It explains the structure and operations within advfs including managing filesets, volumes, and the advantages of logging and cloning for efficient file management. Additionally, the document elaborates on the architecture, components, and transaction management of advfs in Tru64 UNIX version 5.
AdvFS/Advanced File System Ccncepts
- 1. Advanced File System concepts Module 1 Copyright (C) 2008 Hewlett-Packard Development Company, L.P.
- 2. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Objectives • Define the terms: file domains, filesets, and volumes • Describe extent-based storage • Describe logging and the benefits of transactions • Describe at a high level: clones, file striping, trashcan directories • Describe the AdvFS architecture and on-disk format at a high level
- 3. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. File domains and filesets • An AdvFS ‘volume’ represents the actual storage entity within a domain • A file domain is a named set of one or more volumes that provide a shared pool of physical storage • A volume is any mechanism that behaves like a UNIX block device – an entire disk – a disk partition – a logical volume configured with the Logical Storage Manager (LSM) • A fileset represents a portion of the directory hierarchy – follows the logical structure of a traditional UNIX file system – hierarchy of directory names and file names. It's what you mount
- 4. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS characteristics • Within AdvFS, ‘pools of storage’ called ‘domains’ are characteristics that make AdvFS an ‘advanced’ file system • Most other file systems lack the ability to draw storage from a pool shared among multiple filesets • AdvFS goes beyond UFS, by allowing you to create multiple filesets that share a common pool of storage within a defined file domain • A fileset is similar to a file system in the following ways: – you can mount filesets like you can mount file systems – filesets can have quotas enabled – filesets can be backed up. • AdvFS separates the directory layer from the storage
- 5. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS capabilities • Filesets offer features not provided by file systems: – you can clone a fileset and back it up while users are still accessing the original – a fileset can span several disks (volumes) in a file domain
- 6. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Two filesets, one domain, three volumes Domain with 3 volumes Fileset A Fileset B
- 7. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Filesets and partitions Filesets File Domain Volumes (Disk Partitions) Filesets != Partitions
- 8. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Volumes • Volumes are ‘virtual disks’ because they function just as a disk would in less sophisticated file systems • A physical storage building block for a file domain • Any logical UNIX block device – “real" disk partition – hardware RAID logical disk – LSM volume • Administered from /etc/fdmns
- 9. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Displaying a directory under /etc/fdmns # ls -l /etc/fdmns/usr_domain lrwxr-xr-x 1 root system 15 Mar 17 17:56 dsk2g -> /dev/disk/dsk2g
- 10. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Filesets • A file/directory tree mapped to a domain • Created using the command mkfset or through dxadvfs • Mounted like a file system • Administered from /etc/fstab file
- 11. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Filesets Mounting through /etc/fstab # cat /etc/fstab root_domain#root / advfs rw 0 1 usr_domain#usr /usr advfs rw 0 2 local_dmn#alpha_fs /local advfs rw 0 3 local_dmn#users_fs /users advfs rw,userquota 0 3 /proc /proc procfs rw 0 0 /dev/fd /dev/fd fdfs rw 0 0 /backup@tryon /backup nfs rw,bg,noexec,nodev
- 12. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Extent concepts • AdvFS attempts to write each file to disk as a set of contiguous pages • This set of contiguous pages is called an extent • An extent map translates the bitfiles to disk blocks • Pages are added to a file by preallocating one fourth of the file size up to 16 pages each time data is appended to the file • When a file uses only part of the last page, a file fragment is created
- 13. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Extent-based storage logical file extent 1 extent 2 Extent Map Disk Space extent 1 extent 2
- 14. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Displaying extents using the showfile command • Use showfile to view AdvFS details pertaining to an individual file – showfile displays the extent map of each file • Simple files have one extent map • Striped files have an extent map for every stripe segment • The showfile command cannot display attributes for symbolic links or non-AdvFS files • Simple file has one extent map, striped file has more than one extent map
- 15. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Using showfile to display a contiguous file # showfile -x /usr/users/obrien/disktab Id Vol PgSz Pages XtntType Segs SegSz I/O Perf File 596b.8001 1 16 3 simple ** ** async 100% disktab extentMap: 1 pageOff pageCnt vol volBlock blockCnt 0 3 1 576496 48 extentCnt: 1
- 16. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. An extent map displays the following information • pageOff is the starting page number of the extent • pageCnt is the number of 8K pages in the extent • vol is a number indicating which volume within the domain contains this file • volBlock is the starting block number of the extent • blockCnt is the number of blocks in the extent • extentCnt is the number of extents
- 17. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Why logging • Many file system operations involve several widely separated writes to disk – a transaction usually consists of more than one write – crash in between the writes leaves the on-disk file system inconsistent • Fast crash recovery • Improved performance for metadata-intensive operations
- 18. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Logging a transaction • Storage is allocated in the bitfile metadata table (BMT) (log record 1) • Bitfile tag slot is allocated (log record 2) • Directory entry is changed (log record 3) • Transaction is committed (log record 4) • Buffered log records are written to disk • Buffered bitfile pages are written to disk and the log pages are removed
- 19. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Event sequence for logging a transaction Tag Directory "log" tagN Directory 1 2 3 Commit 4 5 6 1 2 3 Log intentions commit record
- 20. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS logging (1 of 2) • AdvFS transaction – modifications to its own metadata (internal structures) – not user file data (unless atomic write data logging has been enabled using chfile -L). • For each transaction, AdvFS: – writes a series of log records describing all changes for an operation to disk and then – performs changes (writes changed blocks to disk)
- 21. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS logging (2 of 2) • In case of crash – on reboot – on-disk log indicates which transactions are complete • File directory to insert a new file name • Fileset tag directory to allocate the new file's tag • Bitfile table to allocate an entry for the new file
- 22. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Cloning a fileset using clonefset (1 of 3) • Lock the master (original) fileset • Create the clone fileset • Copy the tag directory of the master to the clone • Increment the clone count in the master fileset • Set the clone’s cloneID = clone count in the master
- 23. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Cloning a fileset using clonefset (2 of 3) Handling a write to the master involves these steps: • If the cloned bitfile does not already exist in the clone fileset, create a bitfile for the file in the clone fileset • Modify the clone fileset tag dir to reference the new file • Allocate an extent in the new file for the portion being written • Copy the original data to the new extent • Let the write occur to the file in the master
- 24. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Cloning a fileset using clonefset (3 of 3) Domain Application write Backup tool read COW read after clone is created, before any writes first write to a block in the original (master) fileset access to COW write blocks in the cloned fileset
- 25. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Cloning issues • Applications should not be writing to the master when the clone is created – fortunately cloning time is very fast (seconds) due to Copy- On-Write (COW) • A clone is not a backup • A clone is a tool for minimizing down time for a fileset due to backups – make clone of fileset – back up from clone – delete clone
- 26. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. File striping (1 of 2) • The stripe utility directs a zero-length file (a file with no data written to it yet) to be spread evenly across several volumes within a file domain • Existing, nonzero-length files cannot be striped using the stripe utility
- 27. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. File striping (2 of 2) Domain File 1 2 3 4 5 ..
- 28. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Overview of trash cans • Trashcan directories can be attached to one or more directories within the same fileset • Root-user privilege is not required to retrieve files from a trashcan directory • You can restore only the most recently deleted version of a file • You can attach more than one directory to the same trashcan directory; however, if you delete files with identical file names from the attached directories, only the most recently deleted file remains in the trashcan directory • When you delete files in the trashcan directory, they are unrecoverable
- 29. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Trashcans Trashcan Dir rm mv
- 30. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. File domain commands •mkfdmn - Make a file domain •addvol - Add a new volume to the domain •rmvol - Remove a volume from the domain •balance - Distribute storage over the volumes evenly •defragment - Make files contiguous if possible •vfast - Background defragmentation and file balancing (V5.1B)
- 31. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. Fileset commands •mkfset - Make a fileset •chfsets - Change fileset characteristics •clonefset - Make a fileset clone
- 32. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. File commands •migrate - Move a file from one volume to another •stripe - Make an empty striped file •mktrashcan - Make a trashcan directory •chfile - Change file attributes
- 33. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS architecture (1 of 2) File access subsystem (FAS) • Emulates UFS and POSIX file and directory semantics • Uses bitfiles to implement files and directories
- 34. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS architecture (2 of 2) Bitfile access subsystem (BAS) • Manipulates bitfiles: create, open, read, write, add and remove storage • Bitfile: array of 8K pages named via a tag. A tag is a unique identifier within a domain similar to an inode number • Interfaces with buffer cache, VM interface, I/O scheduling • Provides transaction and log management • Provides storage placement and management • Provides domain and fileset management
- 35. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. File access: The Big Picture
- 36. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS architecture overview VFS Block Device Interface File Access Subsystem (FAS) Bitfile Access Subsystem (BAS) VFS operations vnode operations Domains and Volumes Bitfiles Transaction Management
- 37. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS components (1 of 2) • File Access Subsystem (FAS): • POSIX file system layer in AdvFS - translates VFS file system requests into BAS requests • Components: – mount, unmount, initialization – directory operations (lookup, create, delete) – file operations (create, read, write, stat, delete, rename)
- 38. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS components (2 of 2) Bitfile access subsystem (BAS): bitfile layer in AdvFS • Domain operations (create, delete, open, close) • Bitfile set operations (create, delete, clone, open, close) • Bitfile operations (create, delete, open, close, migrate, read, write, add & remove stg) • Transactions management operations (start, stop,fail, pin pg, pin record, lock, recover) • Buffer cache operations (pin & unpin page, ref & deref page, flush bitfile, flush cache, prefetch pages, I/O queuing) • Volume operations (add, remove)
- 39. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS in Tru64 UNIX V5 (1 of 2) • Version 5 of Tru64 UNIX has a new version of the on- disk structure of AdvFS • The previous version of the AdvFS on-disk structure was V3; in Tru64 UNIX V5.0, the AdvFS on-disk structure is version 4
- 40. Copyright (C) 2008 Hewlett-Packard Development Company, L.P. AdvFS in Tru64 UNIX V5 (2 of 2) Additional Features: • Faster directory searches for directories larger than 8K • Quota limits now held in 8-byte fields yielding higher limits • Removal of metadata limitations (such as BMT page 0 restrictions) • Direct I/O allowing I/O direct to the application’s address space (no UBC buffering) • Smooth sync() operations to eliminate the update daemon 30-second system I/O bursts • SMP improvements
- 41. Learning check
- 42. Lab 1