Understanding Storage Systems: Technologies, Trends, and RAID - Prof. Sudhakar Yalamanchil, Papers of Computer Architecture and Organization

Download Understanding Storage Systems: Technologies, Trends, and RAID - Prof. Sudhakar Yalamanchil and more Papers Computer Architecture and Organization in PDF only on Docsity! 1 © Sudhakar Yalamanchili, Georgia Institute of Technology Module: Storage Systems 2ECE 4100/6100 (2) Reading • Storage technologies and trends – Section 7.2 – http://www.storagereview.com/map/lm.cgi/areal_density • Interfacing storage devices to the CPU – Section 7.3 • Redundant arrays of inexpensive disks (RAID) – Section 7.5 – A very good reference for an introductory explanation of the basics a of RAID systems http://www.ecs.umass.edu/ece/koren/architecture/Raid/raid home.html 2 3ECE 4100/6100 (3) Motivation: Who Cares About I/O? • CPU Performance has been doubling every 18 months • I/O system performance grows much more slowly limited by speed of mechanical components – Disk seek speeds have improved on the order of 10%/yr • Amdahl's Law: system speed-up limited by the slowest part! – Time spent in I/O determines application speedup – Speedup limited by 1/s, where s is fraction spent in I/O • Faster CPUs do not lead to corresponding reductions in execution time 4ECE 4100/6100 (4) The I/O Subsystem Processor Main Memory L2 Cache I/O Controller Graphics Network Interface Bridge L1 Cache Processor Memory Bus I/O Bus 5 9ECE 4100/6100 (9) Disk Drive Terminology • Data is recorded on concentric tracks on both sides of a platter – Tracks are organized as fixed size (bytes) sectors • Corresponding tracks on all platters form a cylinder • Data is addressed by three coordinates: cylinder, platter, and sector • Actuator moves (seek) the correct read/write head over the correct sector – Under the control of the controller Actuator Arm HeadPlatter Sector Track Platters 10ECE 4100/6100 (10) Disk Performance • Disk latency = controller overhead + seek time + rotational delay + transfer delay – Seek time and rotational delay are limited by mechanical parts Actuator Arm Head Platters 6 11ECE 4100/6100 (11) Disk Performance • Seek time determined by the current position of the head, i.e., what track is it covering, and the new position of the head • Average rotational delay is time for 0.5 revolutions • Transfer rate is a function of bit density 12ECE 4100/6100 (12) Areal Density • Density in a track is measured in bits per inch (BPI) • Track density is measured as tracks per inch (TPI) • Areal density is bit density per unit area and is BPI x TPI • Check out http://www.storagereview.com/map/lm.cgi/areal_density 7 13ECE 4100/6100 (13) Physical Layout • Older designs had the same number of sectors on all tracks – Outer tracks had lower bit density • Constant bit density recording – Outer tracks have more sectors than the inner tracks – Bit density is not quite constant with inner tracks having higher recording density 14ECE 4100/6100 (14) Disk Performance Model /Trends • Capacity – More recently growth accelerated to 100%/year • Transfer rate – 40%/yr • Rotation + Seek time – Closer to 10%/yr • MB/$ – See Figure 7.4 10 19ECE 4100/6100 (19) Arrays of Inexpensive Disks: Throughput • Data is striped across all disks • Visible performance overhead of drive mechanics is amortized across multiple accesses • Scientific workloads are well suited to such oranizations CPU read request Block 0 Block 1 Block 2 Block 3 20ECE 4100/6100 (20) Arrays of Inexpensive Disks: Request Rate • Consider multiple read requests for small blocks of data • Several I/O requests can be serviced concurrently Multiple CPU read requests 11 21ECE 4100/6100 (21) Reliability of Disk Arrays • The reliability of an array of N disks is lower than the reliability of a single disk – Any single disk failure will cause the array to fail – The array is N times more likely to fail • Use redundant disks to recover from failures – Similar to use of error correcting codes • Overhead – Bandwidth and cost Redundant information 22ECE 4100/6100 (22) Redundant Arrays of Small Disks (RAID) • What size disks should we use? • Smaller disks have the advantages of – Lower cost/MB – Higher MB/volume – Higher MB/watt • Classic paper – “Disk System Architectures for High Performance Computing,” R. Katz, G. Gibson, D. Patterson, Proceedings of the IEEE, vol. 77, no. 12, December 1989 12 23ECE 4100/6100 (23) RAID Level 0 • RAID 0 corresponds to use of striping with no redundancy • Provides the highest performance • Provides the lowest reliability • Frequently used in scientific and supercomputing applications where data throughput is important 0 1 2 3 4 5 6 7 24ECE 4100/6100 (24) RAID Level 1 • The disk array is “mirrored” or “shadowed” in its entirety • Reads can be optimized – Pick the array with smaller queuing and seek times • Performance sacrifice on writes – to both arrays mirrors 15 29ECE 4100/6100 (29) RAID 4 Logical View • Small reads can take place concurrently • Large reads/writes use multiple disks and (for writes) the parity disk • Small writes take four accesses – Parity disk is a bottleneck B0 B4 B8 B12 B1 B5 B9 B13 B2 B6 B10 B14 B2 B7 B11 B15 parity parity parity parity 30ECE 4100/6100 (30) RAID 5 Logical View • Block interleaved distributed parity organization distributes write traffic amongst disks • Best combination of small read and small write performance B0 B4 B8 B12 B1 B5 B9 B13 B2 B6 B10 B14 B2 B7 B11 B15 parity parity parity parity 16 31ECE 4100/6100 (31) Summary Storage: Systems • Extraordinary advances in the densities and packaging • Raw performance limited by drive mechanics • System level organizations such as RAID evolved to continue the growth in performance