Operating Systems Lecture Notes

2014 November 4 • Mass Storage

---

Outline

Disk Operation

• The OS issues a sequence of (start, len) requests to the disk.
• The disk services the requests
  • to satisfy the given requests
  • in a way satisfying various criteria.
• The principle criteria is service time: the elapsed time from request issue to request completion.
  • Minimal service time, or predictable service time, or …

Disk Economics

• Disk service time has three components:
  • Seek time: moving the disk head to the proper cylinder.
  • Rotational latency: moving the sector under the disk head.
  • Transfer time: moving the bits to or from the sector.
• Transfer time is fixed, and rotational latency mostly fixed these days.
• Seek time is the component to influence.

Request Scheduling

• Given a set of disk I-O requests, issue them to provide minimal service times.
  • That is, to minimize seek times.
• The Request (or disk) scheduler is responsible for ordering requests.
• Request scheduling is moving from the OS to the disk controller.

Shortest Seek Time First

• Minimize disk-head movement.
  • The request serviced next is closest to the disk head.
  • Shortest Seek Time First (SSFT) scheduling.
  • Similar to shortest-job next process scheduling.
  • Like SJN, may starve requests; not optimal.

Scan

• The disk head scans from one disk edge to the other.
  • Service the request closest to the disk head in the scan direction: Scan (or Elevator) scheduling.
• No starvation, but uneven service times.
  • One request just before the disk head, one just behind.

Circular Scan

• Scan, but service requests in only one direction.
  • The return scan goes edge to edge without stopping.
• Reduced variability, relatively and absolutely.
• Circular-scan (or C-scan) scheduling.
  • Cylinders are a circular list.

Look-Ahead

• Stop scanning to an edge if there are no requests in that direction.
  • Reverse and scan to the other edge.
• Look scheduling.
  • C-look scheduling: scan back when there's no forward requests.
• In either case, less arm movement, saving time.

Which Scheduler?

• Properties: starvation, performance.
  • SSTF scheduling performs well (small movements) but can starve requests.
  • Other schedulers don't starve requests.
• Performance is demand-dependent.
  • Correlated or random requests.
  • Demand-adaptive schedulers
• Modify SSFT with aging.

Disk Management

• From disk to sectors, tracks and cylinders.
• Higher-level disk structures are helpful.
• The boot sector.
  • ROM-based boot code executes on power-up.
  • Boot code reads disk sector 0 into primary store, and executes it.
  • Boot-sector code takes over initialization.
  • Boot sector written during OS install.

Disk Partitions

• A disk partition is a contiguous sequence of disk cylinders.
  • The rest of the system treats a partition is a virtual disk.
• Partitions improve reliability and performance.
  • Catastrophic disk failures wipe-out only the containing partition(s).
  • Special-purpose access and use can be segregated to a partition.

Swap Space

• Pages get written to swap space when paged out.
• Swap space is a disk partition or a file.
  • “Special” due to performance demands and access patterns.
• Swap space is an array of page slots.
  • Each page slot a small multiple of the page size.

Swap Management

Swap space can be allocated dynamically or statically.
• Static allocation done at process load-time.
  • Safe but wasteful.
• Dynamic allocation done at page-out time.
  • Space-efficient but unsafe.
• The swap map keeps track of allocated-available swap space.

Summary

References

---

This page last modified on 2011 October 9.