Chapter 9 Peripheral Devices Magnetic Disks Basic Disk Operation Performance Parameters and History of Improvement Example disks RAID Redundant Arrays of Inexpensive Disks Improving Reliability Improving Performance Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Magnetic Disk Drives High density and non volatile Densities similar to semiconductor RAM on an inexpensive medium No power required to retain stored information Motion of medium supplies power for sensing More random access than tape direct access Different platters selected electronically Track on platter selected by head movement Cyclic sequential access to data on a track Structured address of data on disk Drive Platter Track cylinder Sector Byte Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Multi Platter Hard Disk Drive Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Multiple Platters Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Simplified View of Disk Track and Sector Organization Comp Sys Design Arch 2nd Ed An integral number of sectors are recorded around a track A sector is the unit of data transfer to or from the disk 2004 Prentice Hall Simplified View of Individual Bits Encoded on a Disk Track Inside tracks are shorter thus have higher densities or fewer words All sectors contain the same number of bytes Inner portions of a platter may have fewer sectors per track multiple zone recording Small areas of the disk are magnetized in different directions Change in magnetization direction is what is detected on read Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Disk Layout Methods Diagram Same number of bits in each zone but more zones as you go to outer cylinders Stallings Computer Organization Architecture Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Typical Hard Disk Sector Organization Serial bit stream has header data error code Header synchronizes sector read and records sector address Data length is usually power of 2 bytes Error detection correction code needed at end Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Winchester Disk Format Seagate ST506 Stallings Computer Organization Architecture Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Disk Formatting Disks are pre formatted with track and sector address written in headers Disk surface defects may cause some sectors to be marked unusable for the software Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Static Disk Characteristics Areal density of bits on surface density 1 bit spacing track spacing Maximum density density on innermost track Unformatted capacity includes header and error control bits Formatted capacity capacity bytes sectors tracks of surfaces sector track surface Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Dynamic Disk Characteristics Seek time time to move heads to cylinder Track to track access time to adjacent track Rotational latency time for correct sector to come under read write head Average access time seek time rotational latency Burst rate maximum transfer bandwidth burst rate Comp Sys Design Arch 2nd Ed revs sec sectors rev bytes sector 2004 Prentice Hall RAID Redundant Array of Inexpensive Independent Disks Reliability and Redundancy for Improved Performance Raid LEVEL 0 Speed improvement only NO reliability improvement Data is striped across several disks so they can be accessed in parallel independent of each other Strips may be physical blocks sectors or other units If there is a reasonable probability that multiple disk requests target different disks they can be accessed in parallel i e overlap seek and rotational delay thus lowering average access time Very good when accessing sequential contiguous data Comp Sys Design Arch 2nd Ed 2004 Prentice Hall Data Mapping For RAID 0 Comp Sys Design Arch 2nd Ed Organization Architecture Stallings Computer 2004 Prentice Hall RAID 1 RAID Level 1 Mirroring writing the exact same data to two different disk drives Reliability Performance If one drive fails the other can be used Can seek on both drives at the same time accept the data from the first one to obtain the data and abort the other request Negatives Must write data to two disks slowest one can determine performance Costly need twice as many disks for the same amount of data Comp Sys Design Arch 2nd Ed 2004 Prentice Hall RAID 0 1 2 Comp Sys Design Arch 2ndOrganization Ed Stallings Computer Architecture 2004 Prentice Hall RAID 2 RAID 2 3 General All disks in the array are used for every I O request Disks are synchronized so each disk head is over the same disk position at every point in time RAID Level 2 Data is striped at the bit or word level across several disks Reliability Additional ECC Error Correcting Codes bits to recover data if one drive fails e g Hamming SECDEC are calculated The ECC bits are placed on additional drives The number is proportional to the log of the number of data disks Performance Data bandwidth improves and correponds to the number of data disks Comp Sys Design Arch 2nd Ed 2004 Prentice Hall RAID 3 RAID Level 3 Striped as in level 2 at the byte level Reliability A simple parity bit is computed for each bit position A single disk improvement over RAID 2 holds the parity bits in the calculated position If a parity error is detected The disk holding the error can be replaced data rewritten The error can be corrected on the fly by performing the proper calculation Performance Overall bandwidth is increased through parallel access of block of data Comp Sys Design Arch 2nd Ed 2004 Prentice Hall RAID 4 RAID Level 4 Similar to level 3 Reliability Blocks are used instead of bytes Data blocks are written read asynchronously independently of one another Blocks are related to each other in that they all contribute to a parity calculation where the parity bits are placed on ECC disk drive in the corresponding block position Performance For reading this is fast due to parallel access to blocks For writing more involved parity computation must be performed since block accesses are independent e g the controller that generates the parity bits must read other blocks to generate the parity Often used in transaction based systems such as in airline reservation systems Comp Sys Design Arch 2nd Ed 2004 Prentice Hall RAID 3 4 Stallings Computer Architecture Comp Sys Design Arch 2ndOrganization Ed 2004 Prentice Hall RAID 5 6 RAID Level 5 Similar to level 4 Reliability Difference from 4 Both data ECC bits are both striped across 3 or more drives Round Robin distribution of ECC bits often used One extra disk is needed