1. Chapter 12 – Mass Storage Structures (Pgs 505-545 )

2. Magnetic Disk Terminology  Transfer Rate: Rate at which data flows from the drive to main memory Positioning Time (time before IO can occur) = Seek Time (time to move the head) + Rotational Latency (time for correct sector to come under head)

3. Disk Construction Block: 512 Bytes (typical)

4. Disk Bus Formats  Always changing and improving  SCSI: Small computer systems interface  ATA: Advance technology attachment  SATA: Serial ATA  EIDE: Enhanced integrated drive electronics  USB: Universal serial bus  FC: Fibre channel

5. The Future of Disks  Surviving for the near future  Slower than persistent RAM  Cheaper than persistent RAM  Older technology tends to remain available if it fulfills a role  e.g., magnetic tapes for second-level backups of financial data (still in use on mainframe systems – cost per bit very low compared to disks)

6. Head Scheduling Concerns Synchronicity: Must requests be processed in order? Average Wait Time: How long is a process typically blocked? Maximum Wait Time: How long could a process be blocked (Important in RTOS) Priority: Is all IO equally important? ** Almost identical to process scheduling **

7. Scheduling Algorithms  First Come, First Served (FCFS)  Fair but not most efficient  Shortest seek time first (SSTF)  Better, but far from optimal  SCAN (Elevator)  Moves back and forth, servicing requests in cylinder order  C-SCAN (Circular)  Same as SCAN but don't service on return trip  LOOK (and also C-LOOK)  Same as SCAN but only go/return as far as needed

8. Choosing One  Often don't have to now  being built into the disk controller  SSTF or LOOK are most common  Simple, decent performance  Sophisticated systems change algorithm based on data properties (like sorting functions in libraries)

9. Disk Management  Much of the low-level management (blocks, ECC, bad-sector replacement) is done by the disk controller  Blocks typically 512K, but 256 and 1024 are sometimes possible  OS uses "Clusters" which are larger than blocks and which function as virtual blocks  Increases efficiency, better supports large files, reduces overhead, maps to page size

10. Bad Blocks  Cheap disks don't do much  Newer/High-end disks handle it via disk controller  Spare sectors exist  Can cause problems with head scheduling  Controllers try to generate replacements on the same cylinder for this reason  Small errors often repairable using ECC

11. RAID  Redundant Array of Independent Disks  Improved reliability  Higher data transfer rate  Less storage (due to redundancy) than if used separately  Often provided as a "unit" with its own independent controller

12. Mirroring  Simplest form of redundancy  Both disks have to fail to lose data  About the most effective method for providing redundancy  Highly expensive with respect to resources  "Independent Failure" is not fully realistic  Bad batches, aging, fire/flood, etc.  NOTHING prevents some data loss due to power-failure

13. Striping  Various Levels  Block-level (most common)  Byte-level  Bit-level

14. RAID Levels 0: Block-striping, no redundancy 1: Complete mirroring 2: Byte-striping, parity disks, ECC for single bit errors 3: Bit-interleaved parity (uses sector parity checks to reduce overhead of RAID 2) 4: Block-interleaved parity (RAID 0 with parity blocks) 5: Distributed block-interleaved parity: Same as 4 but with parity and data intermixed 6: P+Q Scheme: Same as 5, but with improved ECC such as Reed-Solomon coding (tolerates dual disk failures)

15. RAID 0 + 1  RAID 0: Striping, improves performance  RAID 1: Mirroring, improves reliability  Put together they are highly effective  Only half the storage capacity is available due to mirroring  Simple to implement and maintain

16. Selecting a RAID Level  Rebuild Time: How long does it take to react to a failure?  Need for reliability?  Cost effectiveness (mirroring costs more per bit)?  Performance concerns  Type of protection – RAID protects against hardware failures, not against software failures  Properties of file systems – stability, volumes, sizes

17. Other Storage  NVRAM – As a solid state disk or as a write- back cache  Optical Disks (CD/DVD) – Various technologies for CD-ROM, CD-R ("WORM"), CD-RW  Magnetic Tapes: Cartridges (kind of like VHS), slow but very economical

18. Storage Performance Issues  Speed -- Bandwidth & Latency  Effective Bandwidth: Overall transfer rate (including latency)  Sustained Bandwidth: Rate of flow  Average Latency: Time from request until transfer begins  Reliability (Redundancy) – number and kind of failures from which recovery can occur  Cost – price per bit  Lifespan – Will data be required in 20 years?  Access Frequency – How often will data be accessed?

19. To Do:  Work on Assignment 2 (Due next week)  Complete Lab 7 (optional lab)  Read Chapter 12 (pgs 505-545; this lecture)