Application/API
A process calls read, write or an asynchronous interface without knowing controller registers.
PART 3 • MEMORY & STORAGE
Coordinate Devices without Making the CPU Wait
Follow an application request through drivers and controllers, compare programmed, interrupt-driven and DMA transfer, then schedule storage requests explicitly.
BY THE END, YOU CAN
Reason about data movement
- Explain the I/O software stack.
- Compare polling, interrupts and DMA.
- Trace buffering and caching.
- Run FCFS, SSTF, SCAN and C-SCAN.
- Compare foundational RAID levels.
Drivers Convert Generic Requests into Device Commands
Device-independent I/O
The kernel checks permissions, names devices, buffers data and applies common error handling.
Driver
Device-specific software builds commands, manages queues and interprets completion status.
Controller and device
Hardware moves bits and reports completion or error through status registers and interrupts.
Spooling queues complete jobs for a serially shared device such as a printer. Buffering absorbs speed mismatch during transfer. Caching retains reusable data to avoid repeated slow access. The concepts overlap but answer different problems.
Polling, Interrupts and DMA Trade Simplicity for Parallelism
CPU polls and copies
Simple but repeated status checks waste CPU time when devices are slow.
Device announces readiness
The CPU does other work, then saves state and runs a handler on completion.
Controller transfers blocks
The CPU programs source, destination and length; DMA moves bulk data and interrupts after completion.
Measure Head Movement on One Request Queue
For teaching, seek cost is approximated by absolute cylinder movement. SSDs do not have a mechanical head, so these classical policies mainly explain HDD scheduling foundations.
READY
Run one policy
The service path and total movement will appear here.
| Policy | Rule | Benefit | Risk |
|---|---|---|---|
| FCFS | Arrival order | Fair and simple | Large movement |
| SSTF | Nearest request | Lower local seek | Far requests may starve |
| SCAN | Sweep then reverse | Bounded directional service | Middle can wait less than edges |
| C-SCAN | Serve one direction | More uniform waiting | Return movement overhead |
RAID Level Numbers Describe Different Layouts
Striping
High parallel throughput and full capacity, but one disk failure loses the array. It is not redundancy.
Mirroring
Duplicate copies improve read options and tolerate a mirror member failure at roughly half raw capacity.
Distributed parity
Capacity-efficient failure tolerance, but small writes need parity work and rebuild risk must be managed.
RAID is not a backup: it does not protect against accidental deletion, corruption replicated across members, theft or site loss. Backups need separate versions and tested restoration.
Trace One DMA Read
Ten Concept Checks
Answered correctly: 0 of 10
Explain Both Software and Hardware Roles
- What is a driver?
- Define DMA.
- Buffer versus cache?
- What is spooling?
- RAID 0 versus 1?
- Trace an I/O request.
- Compare transfer modes.
- Run SSTF and SCAN.
- Compare RAID levels.
- Explain interrupt handling.
- Why are interrupts expensive?
- Can SSTF starve?
- Why is RAID not backup?
- What changes for SSDs?
- What is memory-mapped I/O?
LEVEL 11 SUMMARY
You Can Trace a Request from Process to Device
- Drivers translate generic operations into controller commands.
- Interrupts avoid continuous polling.
- DMA moves bulk data with limited CPU involvement.
- Disk scheduling trades seek cost, fairness and starvation.
- RAID improves selected availability/performance goals but is not backup.
COURSE CHECKPOINT
Mark complete after calculating one full disk schedule.
Saved only in this browser.