Unformatted text preview: ________number of bytes transferred____________ time from request dispatch to completion of transfer SSD / Flash Memory Block devices (like disk drives) What's different about solidstate "drives"? Access time is nearly uniform (no head movement or disk rotation) Bandwidth is constant Need for uniform distribution of sector usage We'll focus on algorithms for magnetic disk drives. Disk Scheduling Classical algorithms: FCFS: FirstCome FirstServed SSTF: Shortest SeekTime First Elevator types Others: SCAN, CSCAN, LOOK, CLOOK Deadline, Linus Elevator Anticipatory CFQ: Complete Fair Queuing FCFS Requests queued in order received Linux noop scheduler is FCFS Fair: no request is ignored Speed (not) starvation cannot occur Example FCFS request queue: even individual files might be fragmented seeks may take excessive time 2, 16000, 4, 15205, 3 Note: this list of requests is called a "reference string" In Linux, the dispatch queue handles elementary merging Example FCFS queue: 2, 16000, 4, 15205, 3 After merging: (2,3,4), 16000, 15205 Check top of p. 245 Notes about Merging "proceeds" means "comes after&...
View Full Document
- Spring '08
- Operating Systems, Input/output, disk drives, Elevator algorithm, I/O scheduling, request queue, CFQ