1. I/O Management and Disk Scheduling Chapter 11

2. I/O Driver OS module which controls an I/O device hides the device specifics from the above layers in the OS/kernel translates logical I/O into device I/O (logical disk blocks into {track, head, sector}) performs data buffering and scheduling of I/O operations structure: several synchronous entry points (device initialization, queue I/O requests, state control, read/write) and an asynchronous entry point (to handle interrupts)

3. Typical driver structure driver_strategy(request) { if (empty(request-queue)) driver_start(request) else add(request, request-queue) block_current_process; reschedule() } driver_start(request) { current_request= request; start_dma(request); } driver_ioctl(request) { } driver_init() { } driver_interrupt(state) /* asynchronous part */ { if (state==ERROR) && (retries++<MAX) { driver_start(current_request); return; } add_current_process_to_active_queue if (! (empty(request_queue)) driver_start(get_next(request_queue)) }

4. User to Driver Control Flow user kernel read, write, ioctl special file ordinary file File System Buffer Cache block device character device Character queue driver_read/writedriver-strategy

5. I/O Buffering before an I/O request is placed the source/destination of the I/O transfer must be locked in memory I/O buffering: data is copied from user space to kernel buffers which are pinned to memory works for character devices (terminals), network and disks buffer cache: a buffer in main memory for disk sectors character queue: follows the producer/consumer model (characters in the queue are read once) unbuffered I/O to/from disk (block device): VM paging for instance

6. Buffer Cache when an I/O request is made for a sector, the buffer cache is checked first if it is missing from the cache, it is read into the buffer cache from the disk exploits locality of reference as any other cache usually replacements done in chunks (a whole track can be written back at once to minimize seek time) replacement policies are global and controlled by the kernel

7. Replacement policies buffer cache organized like a stack: replace from the bottom LRU: replace the block that has been in the cache longest with no reference to it (on reference a block is moved to the top of the stack) LFU: replace the block with the fewest references (counters which are incremented on reference and blocks move accordingly) frequency-based replacement: define a new section on the top of the stack, counter is unchanged while the block is in the new section

8. Least Recently Used The block that has been in the cache the longest with no reference to it is replaced The cache consists of a stack of blocks Most recently referenced block is on the top of the stack When a block is referenced or brought into the cache, it is placed on the top of the stack

9. Least Frequently Used The block that has experienced the fewest references is replaced A counter is associated with each block Counter is incremented each time block accessed Block with smallest count is selected for replacement Some blocks may be referenced many times in a short period of time and then not needed any more

10. Application-controlled File Caching two-level block replacement: responsibility is split between kernel and user level a global allocation policy performed by the kernel which decides which process will give up a block a block replacement policy decided by the user: kernel provides the candidate block as a hint to the process the process can overrule the kernels choice by suggesting an alternative block the suggested block is replaced by the kernel examples of alternative replacement policy: most-recently used (MRU)

11. Sound kernel-user cooperation oblivious processes should do no worse than under LRU foolish processes should not hurt other processes smart processes should perform better than LRU whenever possible and they should never perform worse if kernel selects block A and user chooses B instead, the kernel swaps the position of A and B in the LRU list and places B in a placeholder which points to A (kernels choice) if the user process misses on B (i.e. he made a bad choice), and B is found in the placeholder, then the block pointed to by the placeholder is chosen (prevents hurting other processes)

12. Disk Performance Parameters To read or write, the disk head must be positioned at the desired track and at the beginning of the desired sector Seek time –time it takes to position the head at the desired track Rotational delay or rotational latency –time its takes for the beginning of the sector to reach the head

13. Disk Performance Parameters Access time –Sum of seek time and rotational delay –The time it takes to get in position to read or write Data transfer occurs as the sector moves under the head

14. Disk I/O Performance disks are at least four orders of magnitude slower than the main memory the performance of disk I/O is vital for the performance of the computer system as a whole disk performance parameters seek time (to position the head at the track): 20 ms rotational delay (to reach the sector): 8.3 ms transfer time: 1-2 MB/sec access time (seek time+ rotational delay) >> transfer time for a sector therefore the order in which sectors are read matters a lot

15. Disk Scheduling Policies Seek time is the reason for differences in performance For a single disk there will be a number of I/O requests If requests are selected randomly, we will get the worst possible performance

16. Disk Scheduling Policies First-in, first-out (FIFO) –Process request sequentially –Fair to all processes –Approaches random scheduling in performance if there are many processes

17. Disk Scheduling Policies Priority –Goal is not to optimize disk use but to meet other objectives –Short batch jobs may have higher priority –Provide good interactive response time

18. Disk Scheduling Policies Last-in, first-out –Good for transaction processing systems The device is given to the most recent user so there should be little arm movement –Possibility of starvation since a job may never regain the head of the line

19. Disk Scheduling Policies Shortest Service Time First –Select the disk I/O request that requires the least movement of the disk arm from its current position –Always choose the minimum Seek time

20. Disk Scheduling Policies SCAN –Arm moves in one direction only, satisfying all outstanding requests until it reaches the last track in that direction –Direction is reversed

21. Disk Scheduling Policies C-SCAN –Restricts scanning to one direction only –When the last track has been visited in one direction, the arm is returned to the opposite end of the disk and the scan begins again

22. Disk Scheduling Policies N-step-SCAN –Segments the disk request queue into subqueues of length N –Subqueues are process one at a time, using SCAN –New requests added to other queue when queue is processed FSCAN –Two queues –One queue is empty for new request

23. Disk Scheduling Policies usually based on the position of the requested sector rather than according to the process priority shortest-service-time-first (SSTF): pick the request that requires the least movement of the head SCAN (back and forth over disk): good distribution C-SCAN(one way with fast return):lower service variability but head may not be moved for a considerable period of time N-step SCAN: scan of N records at a time by breaking the request queue in segments of size at most N FSCAN: uses two subqueues, during a scan one queue is consumed while the other one is produced

24. RAID Redundant Array of Independent Disks (RAID) idea: replace large-capacity disks with multiple smaller-capacity drives to improve the I/O performance RAID is a set of physical disk drives viewed by the OS as a single logical drive data are distributed across physical drives in a way that enables simultaneous access to data from multiple drives redundant disk capacity is used to compensate the increase in the probability of failure due to multiple drives size RAID levels (design architectures)

25. RAID Level 0 does not include redundancy data is stripped across the available disks disk is divided into strips strips are mapped round-robin to consecutive disks a set of consecutive strips that map exactly one strip to each array member is called stripe strip 0 strip 3 strip 2 strip 1 strip 7strip 6strip 5strip 4...

26. RAID Level 1 redundancy achieved by duplicating all the data each logical disk is mapped to two separate physical disks so that every disk has a mirror disk that contains the same data a read can be serviced by either of the two disks which contains the requested data (improved performance over RAID 0 if reads dominate) a write request must be done on both disks but can be done in parallel recovery is simple but cost is high strip 0 strip 1 strip 0 strip 1 strip 3strip 2strip 3strip 2...

27. RAID Levels 2 and 3... parallel access: all disks participate in every I/O request small strips (byte or word size) RAID 2: error correcting code (Hamming) is calculated across corresponding bits on each data disk and stored on log(data) parity disks; necessary only if error rate is high RAID 3: a single redundant disk which keeps the parity bit P(i) = X2(i) + X1(i) + X0(i) in the event of failure, data can be reconstructed but only one request at the time can be satisfied b0 b1 b2P(b) X2(i) = P(i) + X1(i) + X0(i)

28. RAID Levels 4 and 5 strip 0 P(0-2) strip 2 strip 1 P(3-5)strip 5strip 4strip 3 independent access: each disk operates independently, so multiple I/O request can be satisfied in parallel large strips RAID 4: for small writes: 2 reads + 2 writes example: if write performed only on strip 0: P(i) = X2(i) + X1(i) + X01(i) = X2(i) + X1(i) + X0(i) + X0(i) + X0(i) = P(i) + X0(i) + X0(i) RAID 5: parity strips are distributed across all disks

29. UNIX SVR4 I/O