More on FilesCS-4513, D-Term More on File Systems CS-4513 Distributed Computing Systems (Slides include materials from Operating System Concepts, 7 th ed., by Silbershatz, Galvin, & Gagne, Modern Operating Systems, 2 nd ed., by Tanenbaum, and Distributed Systems: Principles & Paradigms, 2 nd ed. By Tanenbaum and Van Steen)

More on FilesCS-4513, D-Term Mapping files to Virtual Memory Instead of “reading” from disk into virtual memory, why not simply use file as the swapping storage for certain VM pages? Called mapping Page tables in kernel point to disk blocks of the file

More on FilesCS-4513, D-Term Memory-Mapped Files Memory-mapped file I/O allows file I/O to be treated as routine memory access by mapping a disk block to a page in memory A file is initially “read” using demand paging. A page- sized portion of the file is read from the file system into a physical page. Subsequent reads/writes to/from the file are treated as ordinary memory accesses. Simplifies file access by allowing application to simple access memory rather than be forced to use read() & write() calls to file system.

More on FilesCS-4513, D-Term Memory-Mapped Files (continued) A tantalizingly attractive notion, but … Cannot use C/C++ pointers within mapped data structure Corrupted data structures likely to persist in file Recovery after a crash is more difficult Don’t really save anything in terms of Programming energy Thought processes Storage space & efficiency

More on FilesCS-4513, D-Term Memory-Mapped Files (continued) Nevertheless, the idea has its uses 1.Simpler implementation of file operations –read(), write() are memory-to-memory operations –seek() is simply changing a pointer, etc… –Called memory-mapped I/O 2.Shared Virtual Memory among processes

More on FilesCS-4513, D-Term Shared Virtual Memory

More on FilesCS-4513, D-Term Shared Virtual Memory (continued) Supported in –Windows XP –Apollo DOMAIN –Linux?? Synchronization is the responsibility of the sharing applications –OS retains no knowledge –Few (if any) synchronization primitives between processes in separate address spaces

More on FilesCS-4513, D-Term Questions?

More on FilesCS-4513, D-Term Problem Question:– –If mean time to failure of a disk drive is 100,000 hours, –and if your system has 100 identical disks, –what is mean time between drive replacement? Answer:– –1000 hours (i.e., days  6 weeks) I.e.:– –You lose 1% of your data every 6 weeks! But don’t worry – you can restore most of it from backup!

More on FilesCS-4513, D-Term Can we do better? Yes, mirrored –Write every block twice, on two separate disks –Mean time between simultaneous failure of both disks is >57,000 years Can we do even better? –E.g., use fewer extra disks? –E.g., get more performance?

More on FilesCS-4513, D-Term RAID – Redundant Array of Inexpensive Disks Distribute a file system intelligently across multiple disks to –Maintain high reliability and availability –Enable fast recovery from failure –Increase performance

More on FilesCS-4513, D-Term “Levels” of RAID Level 0 – non-redundant striping of blocks across disk Level 1 – simple mirroring Level 2 – striping of bytes or bits with ECC Level 3 – Level 2 with parity, not ECC Level 4 – Level 0 with parity block Level 5 – Level 4 with distributed parity blocks

More on FilesCS-4513, D-Term RAID Level 0 – Simple Striping Each stripe is one or a group of contiguous blocks Block/group i is on disk (i mod n) Advantage –Read/write n blocks in parallel; n times bandwidth Disadvantage –No redundancy at all. System MBTF is 1/n disk MBTF! stripe 8 stripe 4 stripe 0 stripe 9 stripe 5 stripe 1 stripe 10 stripe 6 stripe 2 stripe 11 stripe 7 stripe 3

More on FilesCS-4513, D-Term RAID Level 1– Striping and Mirroring Each stripe is written twice Two separate, identical disks Block/group i is on disks (i mod 2n) & (i+n mod 2n) Advantages –Read/write n blocks in parallel; n times bandwidth –Redundancy: System MBTF = (Disk MBTF) 2 at twice the cost –Failed disk can be replaced by copying Disadvantage –A lot of extra disks for much more reliability than we need stripe 8 stripe 4 stripe 0 stripe 9 stripe 5 stripe 1 stripe 10 stripe 6 stripe 2 stripe 11 stripe 7 stripe 3 stripe 8 stripe 4 stripe 0 stripe 9 stripe 5 stripe 1 stripe 10 stripe 6 stripe 2 stripe 11 stripe 7 stripe 3

More on FilesCS-4513, D-Term RAID Levels 2 & 3 Bit- or byte-level striping Requires synchronized disks Highly impractical Requires fancy electronics For ECC calculations Not used; academic interest only See Silbershatz, § (pp )

More on FilesCS-4513, D-Term Observation When a disk or stripe is read incorrectly, we know which one failed! Conclusion: –A simple parity disk can provide very high reliability (unlike simple parity in memory)

More on FilesCS-4513, D-Term RAID Level 4 – Parity Disk parity 0-3 = stripe 0 xor stripe 1 xor stripe 2 xor stripe 3 n stripes plus parity are written/read in parallel If any disk/stripe fails, it can be reconstructed from others –E.g., stripe 1 = stripe 0 xor stripe 2 xor stripe 3 xor parity 0-3 Advantages –n times read bandwidth –System MBTF = (Disk MBTF) 2 at 1/n additional cost –Failed disk can be reconstructed “on-the-fly” (hot swap) –Hot expansion: simply add n + 1 disks all initialized to zeros However –Writing requires read-modify-write of parity stripe  only 1x write bandwidth. stripe 8 stripe 4 stripe 0 stripe 9 stripe 5 stripe 1 stripe 10 stripe 6 stripe 2 stripe 11 stripe 7 stripe 3 parity 8-11 parity 4-7 parity 0-3

More on FilesCS-4513, D-Term RAID Level 5 – Distributed Parity Parity calculation is same as RAID Level 4 Advantages & Disadvantages – Mostly same as RAID Level 4 Additional advantages –avoids beating up on parity disk –Some writes in parallel (if no contention for parity drive) Writing individual stripes (RAID 4 & 5) –Read existing stripe and existing parity –Recompute parity –Write new stripe and new parity stripe 12 stripe 8 stripe 4 stripe 0 parity stripe 9 stripe 5 stripe 1 stripe 13 parity 8-11 stripe 6 stripe 2 stripe 14 stripe 10 parity 4-7 stripe 3 stripe 15 stripe 11 stripe 7 parity 0-3

More on FilesCS-4513, D-Term RAID 4 & 5 Very popular in data centers –Corporate and academic servers Built-in support in Windows XP and Linux –Connect a group of disks to fast SCSI port (320 MB/sec bandwidth) –OS RAID support does the rest! Other RAID variations also available

More on FilesCS-4513, D-Term New Topic

More on FilesCS-4513, D-Term Incomplete Operations Problem – how to protect against disk write operations that don’t finish –Power or CPU failure in the middle of a block –Related series of writes interrupted before all are completed Examples: –Database update of charge and credit –RAID 1, 4, 5 failure between redundant writes

More on FilesCS-4513, D-Term Solution (part 1) – Stable Storage Write everything twice to separate disks Be sure 1 st write does not invalidate previous 2 nd copy RAID 1 is okay; RAID 4/5 not okay! Read blocks back to validate; then report completion Reading both copies If 1 st copy okay, use it – i.e., newest value If 2 nd copy different or bad, update it with 1 st copy If 1 st copy is bad; update it with 2 nd copy – i.e., old value

More on FilesCS-4513, D-Term Stable Storage (continued) Crash recovery Scan disks, compare corresponding blocks If one is bad, replace with good one If both good but different, replace 2 nd with 1 st copy Result:– If 1 st block is good, it contains latest value If not, 2 nd block still contains previous value An abstraction of an atomic disk write of a single block Uninterruptible by power failure, etc.

More on FilesCS-4513, D-Term What about more complex disk operations? E.g., File create operation involves Allocating free blocks Constructing and writing i-node –Possibly multiple i-node blocks Reading and updating directory Update Free list and store back onto disk What if system crashes with the sequence only partly completed? Answer: inconsistent data structures on disk

More on FilesCS-4513, D-Term Solution (Part 2) – Log-Structured File System Make changes to cached copies in memory Collect together all changed blocks Including i-nodes and directory blocks Write to log file (aka journal file) A circular buffer on disk Fast, contiguous write Update log file pointer in stable storage Offline: Play back log file to actually update directories, i-nodes, free list, etc. Update playback pointer in stable storage

More on FilesCS-4513, D-Term Transaction Data Base Systems Similar techniques –Every transaction is recorded in log before recording on disk –Stable storage techniques for managing log pointers –One log exist is confirmed, disk can be updated in place –After crash, replay log to redo disk operations

More on FilesCS-4513, D-Term Journaling File Systems Linux ext3 file system Windows NTFS

More on FilesCS-4513, D-Term Berkeley LFS — a slight variation Everything is written to log i-nodes point to updated blocks in log i-node cache in memory updated whenever i-node is written Cleaner daemon follows behind to compact log Advantages: –LFS is always consistent –LFS performance Much better than Unix file system for small writes At least as good for reads and large writes Tanenbaum, §6.3.8, pp Rosenblum & Ousterhout, Log-structured File System (pdf)Rosenblum & Ousterhout, Log-structured File System (pdf) Note: not same as Linux LFS (large file system)

More on FilesCS-4513, D-Term Example i-node modified blocks a b c Before old i-node old blocks a b c log a b c new blocks new i-node After

More on FilesCS-4513, D-Term Questions? Next Topic