The Big Picture: Where are We Now? I/O System Design Issues - PDF document

The Big Picture: Where are We Now? I/O System Design Issues interrupts Processor Network Performance Processor Expandability Processor Cache Input Input Resilience in the Control face of failure Control Memory Memory Memory - I/O Bus Datapath Datapath Output I/O I/O I/O Output Main Controller Controller Controller Memory Graphics Disk Disk Network Datorteknik F1 bild 1 Datorteknik F1 bild 2 I/O Device Examples I/O Devices ■ Connected to the Backplane bus Device Behavior Partner Data Rate (KB/sec) – Hard disk controllers Keyboard Input Human 0.01 – Graphics adapters Mouse Input Human 0.02 – Serial I/O Line Printer Output Human 1.00 – Sound Cards Floppy disk Storage Machine 50.00 – Network adapters Laser Printer Output Human 100.00 – Virtual Reality ■ Helmet Optical Disk Storage Machine 500.00 ■ Gloves Magnetic Disk Storage Machine 5,000.00 ■ Quake controller Network-LAN Inp or Outp Machine 20 – 1,000.00 Graph. Display Output Human 30,000.00 Datorteknik F1 bild 3 Datorteknik F1 bild 4 I/O System Performance I/O Performance ■ I/O System performance depends on many aspects of the system (“limited by weakest link in the chain”): ■ I/O Bandwidth – The CPU – How much data can we move from A to B/time unit – The memory system: – How many I/O operations can we perform/time unit ■ Internal and external caches ■ Response time ■ Main Memory – The underlying interconnection (buses) – The total time to perform a task – The I/O controller ■ Latency per access – The I/O device ■ Bandwidth – The speed of the I/O software (Operating System) – The efficiency of the software’s use of the I/O devices Datorteknik F1 bild 5 Datorteknik F1 bild 6

Simple Producer-Server Model Throughput versus Response Time Response Time (ms) 300 Producer Queue Server ■ Throughput: – The number of tasks completed by the server in unit time 200 – In order to get the highest possible throughput: ■ The server should never be idle ■ The queue should never be empty 100 ■ Response time: – Begins when a task is placed in the queue – Ends when it is completed by the server – In order to minimize the response time: 20% 40% 60% 80% 100% ■ The queue should be empty Percentage of maximum throughput ■ The server will be idle Datorteknik F1 bild 7 Datorteknik F1 bild 8 Throughput Enhancement Magnetic Disk ■ Purpose: Registers – Long term, nonvolatile storage Server Cache Memory Queue – Large, inexpensive, and slow Disk – Lowest level in the memory hierarchy ■ Two major types: – Floppy disk – Hard disk Producer ■ Both types of disks: Queue Server – Rely on a rotating platter coated with a magnetic surface – Use a moveable read/write head to access the disk ■ In general throughput can be improved by: ■ Advantages of hard disks over floppy disks: – Throwing more hardware at the problem – Platters are more rigid ( metal or glass) so they can be larger – reduces load-related latency – Higher density because it can be controlled more precisely ■ Response time is much harder to reduce: – Higher data rate because it spins faster – Ultimately it is limited by the speed of light (but we’re far from it) – Can incorporate more than one platter Datorteknik F1 bild 9 Datorteknik F1 bild 10 Hard Disk Hard disk 1-10 inches ■ Average Seek Time (average time to move to a track) – 8-12 ms ■ 2-20 Platters – Does not consider locality actual average seek time may be 25% to 2-20 ■ 3600-10000 RPM 33% lower. – (Sum of the time for all possible seek) / (total # of possible seeks) ■ 1-10 Inch Diameter ■ Rotational Latency ■ 500-2000 tracks/surface – 4-8 ms ■ 32-128 sectors/track – Often dominates over Seek Time Sectors ■ Cylinder Track 30 31 32 1 2 3 … – All tracks at one position – Start reading to ring-buffer when track reached ■ Transfer Rate – 2-100 Mb/sec Datorteknik F1 bild 11 Datorteknik F1 bild 12

Typical Numbers of a Magnetic Disk Other Properties Track Sector ■ Storage Size 500Mb-160Gb ■ Cylinder 0, Boot Block ■ Rotational Latency: Cylinder – Most disks rotate at 3,600 to 10000 RPM ■ Partition Information Platter – Approximately 16 ms to 6 ms Head – Usually a Physical Disk is devided into Smaller Partitions per revolution, respectively – An average latency to the desired – File System Information information is halfway around the disk: ■ File System 8 ms at 3600 RPM, 3 ms at 10000 RPM ■ Transfer Time is a function of : – The “data structure” for storing files and folders – Transfer size (usually a sector): 1 KB / sector – Usually Hierarchical – Rotation speed: 3600 RPM to 10000 RPM ■ Folders, sub folders and files – Recording density: bits per inch on a track ■ File types, (program/data-format) and protection bits – Diameter typical diameter ranges from 2.5 to 5.25 in – Typical values: 2 to 100 MB per second Datorteknik F1 bild 13 Datorteknik F1 bild 14 Disk I/O Performance Example Request Rate Service Rate ■ 512 byte sector, rotate at 5400 RPM, advertised seeks is 12 ms, λ µ transfer rate is 4 MB/sec, controller overhead is 1 ms, queue idle Disk Disk so no service time Controller Queue ■ Disk Access Time = Seek time + Rotational Latency + Transfer Processor time + Controller Time + Queueing Delay Disk Disk ■ Disk Access Time = 12 ms + 0.5 / 5400 RPM + 0.5 KB/4 MB/s + 1 Controller ms + 0 Queue ■ Disk Access Time = 12 ms + 0.5 / 90 RPS + 0.125 / 1024 s + 1 ms + 0 ■ Disk Access Time = Seek time + Rotational Latency + Transfer time + Controller Time + Queueing Delay ■ Disk Access Time = 12 ms + 5.5 ms + 0.1 ms + 1 ms + 0 ms ■ Disk Access Time = 18.6 ms ■ If real seeks are 1/3 advertised seeks, then its 10.6 ms, with rotation delay at 50% of the time! Datorteknik F1 bild 15 Datorteknik F1 bild 16 I/O Benchmarks for Magnetic Disks SCSI/EIDE I/O Bus ■ Supercomputer application: – Large-scale scientific problems => large files ■ SCSI – One large read and many small writes to snapshot computation – General purpose interface for HDs, Scanners, Streamers, etc – Data Rate: MB/second between memory and disk – Both synchronous/asynchronous operation ■ Transaction processing: – 160MB/sec (Ultra160) – Examples: Airline reservations systems and bank ATMs – 15 Devices/Bus Mastering/Self Selection Arbitration(Wide) – Small changes to large sahred software ■ EIDE – I/O Rate: No. disk accesses / second given upper limit for latency ■ File system: – HDs and HD like devices (CD-ROMs etc) – Measurements of UNIX file systems in an engineering environment: – Synchronous ■ 80% of accesses are to files less than 10 KB – 100MB/sec (ATA100) ■ 90% of all file accesses are to data with sequential addresses on the disk – 2 Devices on each controller ■ 67% of the accesses are reads, 27% writes, 6% read-write – HD controller on Interface – I/O Rate & Latency: No. disk accesses /second and response time Datorteknik F1 bild 17 Datorteknik F1 bild 18