Summary of Computer I/O and Storage Systems
Computer I/O & Storage Systems: A Student's Guide
Introduction
Storage and I/O are central to computer systems: they move data between devices, store information persistently, and determine how fast and reliable applications run. This chapter summarizes device behavior, performance metrics, interconnection methods, I/O programming models, and reliability strategies such as RAID.
Definition: I/O device — A hardware component used for input, output, or storage that communicates with the CPU and memory via controllers and buses.
1. I/O System Characteristics
Break complex ideas into smaller parts:
Behaviour and Partners
- Behaviour: input, output, or storage.
- Partner: human (e.g., keyboard, display) or machine (e.g., disk, network card).
- Data rate: measured in bytes/sec or transfers/sec.
Performance and Dependability
- Latency (response time): time to complete an individual I/O operation.
- Throughput (bandwidth): amount of data processed per unit time.
- Desktops/embedded systems prioritize low latency and device diversity. Servers prioritize high throughput and expandability.
Definition: Latency — The elapsed time from when an I/O request is issued until it completes.
Definition: Throughput — The sustained rate at which data or operations are completed, typically in MB/s or ops/sec.
2. Dependability Measures
- Fault: component failure that may or may not lead to system failure.
- MTTF (Mean Time To Failure): expected operating time before a failure occurs.
- MTTR (Mean Time To Repair): expected time to restore service after a failure.
- MTBF (Mean Time Between Failures): $\text{MTBF} = \text{MTTF} + \text{MTTR}$.
- Availability: $\text{Availability} = \dfrac{\text{MTTF}}{\text{MTTF} + \text{MTTR}}$.
Improving availability:
- Increase MTTF: fault avoidance, fault tolerance, fault forecasting.
- Decrease MTTR: better diagnosis tools, streamlined repair processes.
3. Disk Storage: Organization and Access
Disk sectors and error handling
- A sector contains: Sector ID, Data (commonly 512 B or proposed 4096 B), Error Correcting Code (ECC), sync fields, and gaps.
- ECC hides defects and recording errors.
Definition: Sector — The smallest addressable unit on a disk containing data and metadata required for access and integrity.
Components of disk access time
- Queueing delay: waiting for other pending requests.
- Seek time: moving the read/write head to the target track.
- Rotational latency: waiting for the sector to spin under the head.
- Transfer time: moving data across the interface.
- Controller overhead: processing time on the disk controller.
Example calculation (illustrative):
- For 512 B sector, 15,000 rpm, average seek $4,\text{ms}$, controller overhead $0.2,\text{ms}$, transfer rate $100,\text{MB/s}$:
- Rotational period $=\dfrac{60}{15000},\text{s} = 4,\text{ms}$ so average rotational latency $=2,\text{ms}$.
- Transfer time $=\dfrac{512,\text{B}}{100\times10^6,\text{B/s}} = 0.005,\text{ms}$.
- Average read time $=4,\text{ms} + 2,\text{ms} + 0.005,\text{ms} + 0.2,\text{ms} \approx 6.2,\text{ms}$.
Practical notes
- Manufacturers quote average seek time over all seeks; OS scheduling and locality typically reduce real average seeks.
- Smart disk controllers do logical-to-physical mapping and use caches to prefetch data.
4. Flash Storage
- Nonvolatile semiconductor-based storage, faster and more robust than disks but more expensive per GB.
- Two main types:
- NOR flash: random read/write, used for instruction memory in embedded systems.
- NAND flash: denser, block-at-a-time access, used in USB
Zaregistruj se pro celé shrnutíAlready have an account? Sign in
Storage and I/O Essentials
Klíčové pojmy: I/O characterized by latency (response time) and throughput (bandwidth)., Availability = MTTF / (MTTF + MTTR); improve by increasing MTTF or reducing MTTR., Disk access time = queueing + seek + rotational latency + transfer + controller overhead., Flash: NOR = random access, NAND = block access; use wear leveling for longevity., Buses: processor-memory (short, fast) vs I/O buses (longer, standardized)., I/O models: polling wastes CPU, interrupts are asynchronous, DMA moves data without CPU involvement., Ensure cache coherence for DMA by flushing/invalidation or using non-cacheable memory., RAID levels: RAID 0 (striping), RAID 1 (mirroring), RAID 5 (distributed parity), RAID 6 (dual parity)., Benchmarks (TPC, SPEC SFS, SPEC Web) measure throughput and response under workloads., Design: find weakest link for throughput; use queuing models under load., Modern drives internal remapping and caching can make OS-level disk scheduling less effective., Peak theoretical I/O rates rarely equal sustained system throughput due to other component limits.