Input and Output Devices + I/O Device Characteristics: + Terminal. + One character (8 bits of data or control function) is sent at a time, one interrupt per character. + 10-1900 characters per second. (300-19200 baud. about 10 bits per character transmitted) (68400 available soon) + Keyboard and display are independent in most systems (no automatic echo, full duplex). (discuss full du- plex, half duplex - show diagram of idea.) + Used to be handled with one interrupt per character, but often DMA nowadays. - depends on the machine. PDP-11 and VAX have one interrupt per character. Mainframes have controllers to handle characters. + Line Printer. + Fixed 132-character records (printable characters on- ly). + The first character of each line is a control func- tion like ‘‘normal’’, ‘‘skip to next page’’, etc. (On some machines, can send those characters. E.g. "1" is skip page, "2" is skip half page, "4" is skip 1/4 page, etc.) — 13.1 — + Typical high speed is 2000 lines per minute, but varies widely depending on price of printer. + Most common technology is print chain - a belt of letters. Hammer strikes for each column as character goes by. + Alternate technology is a drum, which rotates in vetical plane. Hammers are behind the drum, and hit character as it goes by. + Some huge laser printers (e.g. IBM) can do several times that speed. Costly (e.g. $300,000). At SLAC, they replaced five high speed line printers with one laser printer. + Could empty a box of printer paper in 10 minutes. Operator load time for paper was a significant problem. + Raster Printer. + Page consists of an array of dots, called ‘‘pixels’’. (picture elements). Typical density is about 300-1200 pixels per linear inch (about 8 million pix- els per page). Professional equipment up to 1200 pixels/inch. + Each dot can be made black or white individually. + Printing rate is typically 5-30 pages per minute, for desk top size units. + Since each pixel is individually controllable, can — 13.2 — create new fonts and draw pictures. + Most raster printers use lasers and xerographic tech- niques, or ink jet. + In a typical color laser printer, the paper makes four passes around the imaging drum to pick up each successive layer of toner (one pass each for cyan, magenta, yellow and black.) So print speed is 1/4 of B/W print rate. Expensive ones (e.g. new Xerox) use 4 lasers (for cyan, magenta, yel- low, black toner). + Newer color "laser" printers use LED arrays in- stead of lasers for the print heads. The engine can contain four print elements (one per color) arranged in a line, so that the paper can move along a straight path to pick up all four colors in a single pass. + Alternate approach is ink-jet printer. Squirts out small droplets of ink. Ink drops can be generated either by heat (by boiling out a small amount) or by pressure on the cartridge. + Hardest problem in printer technology: paper mechan- ics. + Displays + CRT - Cathode Ray Tube + Phosphor coated screen - glows when electron beam strikes it. — 13.3 — + High voltage generates electron beam. Electro- magnetic field deflects the beam onto specific pixels. + Color is obtained by having different pixels for each color. Beam only writes to appropriate sub- pixels. + Liquid Crystal + Each pixel is a liquid crystal which can be ori- ented by an applied electric field. + Each pixel is turned on/off by either a column and row select ("passive matrix") or by an indi- vidual transistor ("active matrix" or "thin film transistor"). + Light is provided by a backlight. Either passed or blocked depending on whether pixel is on or off. 1.2 to 2.4 watts. (total power for LCD is around 2-5 watts). Backlight can be flourescent or (new) LED. + Colors provided by having sub-pixels and color filters over each. + Manufacturing very difficult - like silicon chip - yield problem. Thus very expensive. + Response time: 150ms for passive, 40ms for ac- tive. (now under 10ms). + Newer Technologies: + Organic Light Emitting Diodes - emit light when cur- rent applied. Short lifetime for certain colors. — 13.4 — + Plasma - plasma radiates, and causes phosphor to emit light. Short lifetime originally - currently longer. Separate pixel/color. + Field Emitter Displays - electron gun behind each pixel. + Electroluminescent - emit light when current is passed through them. + Electromechanical - e.g. TI Mirror Chip. + Retinal scanning lasers- write the image on your retina. + Reel to Reel Tape. + 9 tracks, 1/2" wide by 2400 feet long. Also comes in 600’ and 1200’ reels. (Old tapes were 7 track. No longer used) + >From 800 to 6250 bytes per inch (bpi). (Old tapes were also 200bpi and 556 bpi) Max capacity of 6250 tapes is about 180 Mbytes. + show diagram of tape + Technology is basically the same as home tape. Tape base is coated with thin magnetic coating. Head mag- netizes small bits on tape surface. + Variable length records on some machines (about 1-32000 bytes). + Inter-record gap is about .6 inches. + Tape moves at 20-200 inches per second. Max data — 13.5 — rate is about 1.25 Mbytes/sec (IBM 3400 series drives). High speed drives use air columns to avoid tape reel inertia. + Can read or write, but cannot write in middle. Can skip records. + Tape is normally read only one direction. On some machines (e.g. IBM 370), can read tapes backwards. (Not easy to do - have to do system calls in assem- bler.) + Tapes are DMA devices, not one interrupt per charac- ter. + High performance tape drives are expensive - e.g. $30000, plus the cost of a controller. + IBM tapes have some standard formats. Can be la- beled. - i.e. every file has a standard header. + Minimizes probability of overwriting wrong tape. + Disks and tapes read and write blocks of information rather than single bytes: + Storage efficiency: For tapes written at 1600 bpi, 80-byte records use .05 inch, gaps use .6 inch, tape is used 1/12. However, 8000-byte records use 5 inches so gaps are only about 11% of the tape. + Example: HP 1/2" tape (rack mount), 1993: + 6250bpi, 140MB total, mtbf 22,400hrs, 125 inch/per sec, 781KB/sec, power 170-120 watts, rewind time 90 — 13.6 — sec (2400’). + Newer tapes (IBM 3480 type) are cartridges (about 6" x 6"). Hold about same capacity as 2400’ tape. 18 track. 220MBytes. Cost (end user, 1990, IBM - $95,000) + Example: 1992: Cranel: + 18 tracks, 3480 compatible, 540 foot tape, .5 inch wide, cartridge is 1" x 4.3" x 5", MTBF 15000 hrs, tape speed - 1 meter/sec, transfer: 3MB/sec, 18 tracks, cartridge capacity 200MB + Used in automated tape libraries. + DECtape - "randomly" readable and writeable. Used on DEC minicomputers, as cheap addressable storage. Now obsolete. 1960-70s. + DAT tape - 4mm tape in DAT cartridge, (smaller than audio casette). Capacity about 1 - 2Gigabytes. + Uses 3 level error correction - errors are 1 in 10**15. + Uses embedded subcodes to find files and tracks. Tape is blocked in 512KB blocks. Can fast forward to appro- priate block. Data is also organized into ‘groups’ of 126632 bytes. Each group contains 22 logical data frames of fixed capacity. + There is a mode which permits random read/write. Must preformat the tape into frames. Called ‘update in — 13.7 — place.’ + Head/tape speed of 123 inches/second. + Track angle: 6 degrees. + Uses 2 read heads and 2 write heads. Write herringbone pattern of tracks, which can overlap without interfer- ence. + Example, 1993: Hewlett-Packard: + 2GB, 183KB/sec, ave seek 30 sec, 3.9watts, mtbf: 50,000hrs. One version available with data compres- sion, capacity factor 2-4x. + Example 1992: Cranel: capacity: 1.3Gbytes, 60meters long, 183KByte/sec max sustained, 1.5Mbytes/sec burst, MTBF 40000 hrs., search at 200 times faster than normal. Errors: 1/10**15. + Exabyte tape - 8mm tape in cartridge. Holds 2.5 - 5.0 GBytes. + Helican scan device - writes from high speed rotating drum to slow moving tape. + 1995 model - 5 gigabytes, 500 Kbytes/second, 9 tracks. tape is 8mm. Up to 75Mbytes/square inch. tape is .5 inches/second, (150 inches/ second relative to head), 246Kb sustained data rate. + Higher speeds could be obtained by faster drum rotation and higher linear bit densities. 3Mbyte/second feasible. + Example: Cranel: transfer: 1.5Mbyte/sec peak, 246Kbyte/sec sustained, 34200 bits/inch, 819 tracks/inch, — 13.8 — 35 million bits/sq in., .429 inches per second of tape speed, rotor 1800 rpm (effective tape speed 150in/sec), rewind: 75 times normal, file search speed: 10 times nor- mal, MTBF: 20000 hrs, error rate: 1/10**13. + DLT Tape (Digital Linear Tape (quantum 2004)) + Records data in serpentine pattern - along length of tape, then reverse and back. + Tape cartridge is about 4.1"x4.1"x1" + Tape length is about 2000’ + Tape width is about .5" + Purported tape life is >30 years ("less than 10% loss in de- magnetization at 20C and 40% non-condensing humidity). (hah!) + Media durability: 1M passes. + Claimed reliability: 250,000 hours MTBF + Capacity is up to 300GB/tape (before compression). Compres- sion yields 2X. + Maximum tape transfer rate (depending on drive) is up to 36MB/sec. Burst transfer rate up to 200MB/sec. + Typical track density is around 1490 tpi (640 tracks serial serpentine) + Purported error rate - uncorrected 1/10**17, undetected 1/20**27. + Recording density - 233Kbits/inch. + Media durability- 1*10**6 passes + Power (quantum) - 32 watts — 13.9 — + Average file access time: 79sec. (quantum) + Low end version: capacity 40GB (uncompressed), 3MB/sec, 168 tracks, 123KBits/inch, 336 tracks/inch, MTBF 200,000 hours, 15 watts) + Other Tapes: + Variety of other tape formats available. + Two general formats: + Linear tracks - along length of tape. + Helical scan- tracks are diagonal along tape. + Variety of sizes of tape, lengths of tape. + Variety of tape reel sizes; some are 2 reel cartridges. + Tape Issues: + Tapes deteriorate over time. + Variety of formats - mutually incompatible + Tapes are slow (usually 1-3MB/sec, except for very high end). + Formats become obsolete; may not be able to read in fu- ture. + High end tape drives are expensive. + Tapes are cheapest way to save massive amounts of data. + (Hard) Disk. + Draw picture of spindle, platters, read/write-arm. — 13.10 — Mention spaces between platters. Show prop. + Technology similar to tape. Heads float over disk sur- face, at a distance about 1 micron (millionth of a meter) - considerably less than width of human hair. + If disk becomes contaminated, head can crash. + Number of platters (2 surfaces per platter, sometimes ex- cept top and bottom), cylinders, tracks, sectors, bytes per track, etc. are all variable. (Define terms) + Data can be written in variable size or fixed size blocks (sectors). IBM style high end disks permit variable size blocks (called "count key data (CKD) disks"). Most other machines use fixed sectors. (almost always 512 bytes) + Material written looks like tape block: inter-block gap, key field, control info (physical block address, record number, error correction, key length, byte count), plus actual data. Overhead for variable blocks is about 50-100 bytes, plus IRG. Smaller overhead for fixed blocks. + Most disks used to be removable. Now no hard disks are removable (except ZIP (100, 200, 750MB), JAZZ (1-2GB) disk- IOMEGA, Syquest, etc.). (last "major" removable regular disk was IBM 3330.) + Floppy Disk - 1.44MB. 60KB/sec. + To do a disk I/O (internal): + Seek: (give cylinder and track address) - move read/write — 13.11 — head to correct track. + Set Sector - Wait for disk to rotate desired sector into position. (Often channel can let go of disk while this happens. Interrupt generated when right place is reached.) Rotational Latency + Read or write sector while it spins by. + Note that in modern disk controllers, all of this is built in- you issue the I/O, with a block number (linear between 0 and size of disk) and the seek/set sector/read-write are all done invisibly to the CPU. + The embedded code ("microcode") in the disk controller is huge. 100K- 200K lines of assembler. + RPS Miss + Problem with older disks is that path from disk head to CPU had to be available to do read/write. + So, if the path wasn’t available when the sector ro- tated under the head, the I/O could not start, and an RPS Miss occurred. (RPS=Rotational Position Sens- ing). Needed to wait for another entire rotation. + Problem solved in modern disks by buffering within the controller. Data is held in buffer and trans- ferred when path is ready. + Disk Characteristics (2007) Typical sizes: 5.25", 3.5", 2.5", 1.8", 1" — 13.12 — Typical (high end) specifications, 2007: 2.5" - up to 160GB 3.5" - up to 750GB+ 5.25" - up to ?; largely obsolete. sector size - 512 bytes areal density - up to 133Gbit/sq. in. (in research, ~400 Gbit/sq. in.) 3600-15000 RPM ave seek times - 3-5 ms, depending on performance track to track seek - .3ms-2ms, max seek 8-24ms media transfer rate- up to 650MBit/sec sustained data transfer rate - up to about 98MB/sec instantaneous data transfer rate: up to about 320MB/sec (Ultra SCSI 320) bit error rates: recoverable <1/10**12, nonrecoverable: <1/10**16 reliability(??) - >=300,000 hours mtbf (some claim 10**6) reliability - min 50K stop/start cycles start up time - 1s to 10s buffer caches - 64K to 4M power - e.g. read/write - 5-10W, idle 5W, standby .5W, sleep .5W power - e.g. seek 2.5, r/w 2.5, idle 1.0, standby .3, sleep .1W power - 5-15W for desktop type power - e.g. seek 12.9W, R/W 9W, idle 6.6W, standby 1.75W. 512 byte blocks (still) — 13.13 — buffer size- 128KB - 8MB platters: 1-4, 2 surfaces/platter First Disk: RAMAC, 1956. 5MBytes. 24" diameter. Over 1 ton. Rates of increase: originally around 35%/year. 1998-2002 - around 100%/year currently 30-40%/year. Seagate 3.5" Barracuda (2007) up to 750GB (8 heads, 4 platters) 7200 RPM 2, 8, 16 MB Cache up to 300MB/sec 512 bytes/sector 50000 start/stops non recoverable errors: 1/10**14 annualized failure rate: .34% ave pwr - seek - 12.6W operating - 13.0W idle: 9.3W sound: idle 27db, seek: 30db temp range: 0-60C operating, -40 - 70C non operating max shock, operating - 60G, nonoperating: 350G — 13.14 — Seagate 2.5" Momentus up to 120GB (2 platters, 4 heads) 5400 RPM 42MB/sec internal, 150MB/sec external 8MB cache ave seek 12.5ms 512 byte sectors 600000 load/unload cycles 1/10**14 non-recoverable errors power: approx 2 watts operating, .8 watts standby operating: 5-55C nonop -40 to 70C shock, operating: 250G, nonop 900G 24db idle, 29db seek Seagate 3.5" Cheetah 15000 RPM up to 150GB (4 platters, 8 heads) 320GB/sec (external) ave seek: 3.5/4.0 ms track-to-track .2/.4 ms (R/W) internal transfer rate 685 to 1142Mbits/sec sustained 58-96MB/sec 8MB cache MTBF: 1400000 hours. power: ~15 watts operating, ~10watts idle sound: idle 36DB — 13.15 — Maxtor Atlas 15K II (2004) 147GB/4 platters/8 heads 512 byte sectors. Ultra320 SCSI ave seek 3-3.8ms max seek: 9ms track to track seek: .3/.5 ms 15000 rpm sustained data rate: 98MB/sec 8MB cache max 50000 start/stop recoverable - less than 10 per 10**12 unrecoverable - less than 1 per 10**15 temp range 5C to 55C altitude: -1000 to 10000 noise - up to 36db power - 9W to 14W (1-4 platters) + Toshiba 1.8" disk (2001): 5GB/disk 22.4 Gbpsi IBM Microdrive (2000) 170MB-1G, 3600RPM or 4500RPM Disk diameter: 27.4mm Overall size: 43x36x5mm (roughly 1.4" platter) — 13.16 — Areal density: up to 15.2Gb/sq.in. Ave seek time: 12ms Transfer rate: up to 4.2MB/sec. Power: spin 260mA, R/W 250mA, idle 140mA, standby 20mA. (3.3v, 5.0v) Weight: 16g Used in Digital Cameras, portable PCs, etc. Maxtor Atlas 10K V (2004) up to 300GB, 3.5" Uses U320 and U160 (ultra-scsi) interface. 37.25GB/surface, up to 4 platters, 2 surfaces/platter Areal density - up to ?GB/sq. in. Ave read: 4.5ms track-track seek: 0.3ms full stroke: 11ms 10000 RPM internal data rate: 350 to 622Mb/sec sustained throughput: 32 to 55 MB/sec 8 MB buffer errors: < 10/10**12 recoverable, <10/10**15 unrecoverable. 8-11 watts idle power Maxtor DiamondMax 10 (2004) up to 300MB, 3.5", 7200 RPM ave seek: <9.0 ms drive ready time: 7.5ms — 13.17 — 16MB buffer >50000 start/stop cycles Maxtor DiamondMax 16 (2004) 3.5", up to 160GB, 5400 RPM, 2MB cache, 133MB/sec max data rate ave seek: <12.6ms Logical CHS: 16383/16/63 >50,000 start/stop Maxtor DiamondMaxPlus D740X (2002) up to 80GB, 7200 RPM, 3.5" seeks: track/track: 0.8ms, ave: 8.5ms, full: 17.8ms drive ready time: 15ms max data rate: 133MB/sec, media: 54.2MB/sec 2MB buffer >50000 start/stop cycles errors: < 1/10**14 non recoverable. power: spinup: 24W, seek: 12.5W, R/W: 8W, idle 7.4W, standby: 0.8W IBM Travelstar (2000) up to 32GB, media transfer rate up to 228Mbits/sec, 4 disks in 12.5mm up to 17.1Gbits/sq.in. 10GB/platter. 2.5" disk. up to 22784 cylinders. up to 451KB per inch, 38000 tracks/inch, — 13.18 — 4200 RPM, ave seek 12ms, media transfer 109-203Mbits/sec.k track-track seek 2.5ms, max seek 23ms. power: start 4.7W, seek 2.3W, R 2W, write 2.1W, "perfor- mance idle" 1.85W, "active idle" .85W, "low power idle" .65W, standby .25W, sleep 0.1W. Firmware is around 125-175KB. IBM Deskstar 75GXP (2000) 3.5" up to 75GB 7200 RPM sustained transfer rate: 37MB/sec ave seek 8.5ms cylinders 27724 max areal density 11Gbit/in. max density 391kbit/inch track density: 28350/in. max media transfer rate 444Mbits/sec max interface transfer rate 100MB/sec. seek: ave 8.5, 1 track 1.2ms, max 15ms. up to 40000 start/stop error rate (unrecoverable) 1/10**13 + Characteristis of Modern disks + Smaller - typically 3.5". ; 5.25 are rare. 2.5" and 1.8" used in portables. PC card with 1.8". 1.0" used for — 13.19 — cameras, Ipods, etc. + Most or all now contain semiconductor storage as buffer or cache. + Defects remapped within the controller. + Alternate tracks - entire track remapped to else- where. + Alternate sector- at end of track or cylinder + In-line sector sparing- spare sector at end of track, but renumber the sectors, so that blocks look consec- utive. + Addressing is now logical, using linear address space. Mapped by the controller. + Originally wrote at constant freq., i.e. same number of bits on all tracks. + Currently, can write different number of bits per track. Controller keeps track of this. More bits on outer tracks. This increases capacity by 35-50%. Organized in zones. + Current rotation rates between 3600RPM - 15000 RPM. + MTBF has gone from 30,000 hours to 300,000 (1.4M?) hours.(?) + Extensive error correction - correct multiple long error bursts. (Reed-solomon codes.) + Over time, access time has decreased slightly. Maybe 10%/year. + Density increasing by: factor of 10 over 8 years (Hospodor/Hoagland) to 60% per year (EET 8/26/96); trans- — 13.20 — fer rates increasing 30-40%/year. + Power consumption increasingly important. Usually goes as about the square of the speed. + Note that fast disks are noisy and hot. + Standard interfaces such as SCSI, SCSI-2, etc. (SCSI-1 runs at 5MB/sec; newer ones run at 10-320MB/Sec. SCSI can address 8-16 devices per cable. Cables are short. 1.5 to 25 meters depending on which type of SCSI.) + Tell about disks walking across floor, heads flying, pressur- ization, disk crash. + Solid State Disk (SSD) + E.g. Quantum- 268MB to 3.2GB, >9000 I/Os/sec, >30MB/sec, 11-18 watts. + Drums + Show diagram of drum + Idea of drum was to avoid seek time, which disk had. + Drums had small capacity (very little surface area, com- pared to disks) and very high cost (one head per track). + Drums haven’t been manufactured for about 20 years. + But you will still hear about them. + Used to be used for paging. Now disks are used. Called paging drum. + Optical disk (CD): — 13.21 — + Like compact disk, but read/write, or write-once + Example (1993) Maxtor, erasable: + 644 MB (formatted - standard disk), 1020 MB (format- ted, zoned constant angular velocity recording method), ave seek 35ms, ave latency 13.6ms, transfer rate 6.8MB/sec (ZCAV 10MB/sec), 35 watts, 50000hrs mtbf, technology: magneto optic. + Example: Cranel- 650 Mbytes@1K sectors, - 1992 min seek: 22ms, ave seek 95ms, max seek: 185ms, 25ms rotation (2400 rpm), data transfer: read max sustained: 680Kbytes/sec, writes 340Kbytes/sec, burst 1.2Mbytes/sec. 18751 tracks/surface, 15875 tracks/inch. Errors 1/10**14. + Performance - a 1X CD-ROM drive is 150KB/sec. + Maximum physical rotation rate permits 12X. Some rated much higher, but not for full disk. + Traditional CD-ROM drive uses constant linear veloci- ty, by which the bit rate is constant. (This is what you want for audio compact disk.) + More recently, CD-ROM drives use constant angular ve- locity, which means data rate will vary. Track starts at inside of disk. Max rate applies only at end of track (outside of disk). + Sometimes, the disk will slow down at the end of the track, because the bit rate becomes too fast. + Newer high-speed disks read several tracks at once, — 13.22 — to produce high data rates. + Recording Technology + Manufactured - pressed, physical pits + Magneto-optical - change small magnetic domain (usu- ally by heating, with applied external field). Read by detecting differing polarization due to different magnetic field. + Phase change- chemical change due to write laser. + Note that read laser much weaker than write laser. + Digital Video Disk (DVD) + Capacity 4.5GB, 9GB, 18GB (approx) + One sided/2 sided, 1 layer/2 layer + Size similar to existing compact disk. + Higher densities available with blue or green (rather than red) lasers. + thickness 0.6mmx2 beam spot size 1.32um light 650nm + working distance 1.7mm, max transfer rate 10.08Mb/s + Blue Ray and HD Disks + Blue Ray: 25GB/layer. + thickness 0.1mm+1.1mm beam spot size 0.58um light 405nm + working distance 0.5mm, max transfer rate 40Mb/s + HD Disk: 15GB/layer — 13.23 — + Access efficiency: on disk it typically takes 5-15ms over- head before transfer begins. + CPU overhead - around 3K-25K instructions. + Seek time - ave. 2-12ms, range 0ms - 30ms. (note seek averages only 1/3 of disk surface.) + Rotational latency - 2 - 8.33ms ave. + Data transfer. + total is about 5-25ms. + Thus most software that deals with disks and other I/O de- vices attempts to process information in large blocks (usual- ly sequentially). + Device Interconnection + In small systems, CPU connects directly to device con- troller, which drives device. Controller may be built into device. + Multiple devices can be attached to SCSI bus. + In IBM mainframe systems, there are channels. Channels connect to storage control units. Storage controllers connect to string controllers. String controllers have a number of disks on them. + There can be multiple paths from CPU to disk. + Devices can be shared among CPUs at the level of the storage controller or string controller. + NAS and SAN — 13.24 — + NAS - network attached storage. Storage attaches to lo- cal area network (e.g. ethernet). Provides "file" inter- face. Low to midrange product. + SAN - storage area network. Separate(?) network contain- ing storage. "block level" interface. mid-high end product. + Storage networking Industry Associaton (SNIA) - working on standards, so NAS, SAN are standard and can interoper- ate and attach to many types of systems. + Storage Service Providers + Storage provided by third party. Often connects over in- ternet, or via dedicated cables to provider. Expensive.