CDA-4101 Lecture 5 Notes

There is, and always has been, far more data to be stored than is economically feasible in primary memory (RAM). There is also the volatility of RAM that is a problem. Secondary store is far slower, but far larger than primary memory, and also more permanent (but nothing lasts forever.)

• The same idea of using caches to aid RAM's access time can be applied on many levels
• we saw this with the various levels of cache in the last lecture
• we also saw this in the registers that a CPU has
• we now look at all the storage devices and method below the RAM level i.e., secondary storage

• the speed gap between RAM and Magnetic disks is greater than the gaps between the other levels of the hierarchy.

note that and floppy disk drives are also "Magnetic Disks" aluminum platter (disk) with magnetized coating (layer of very small magnetic particles) coated on both sides, so two surfaces contain data induction coil (the main part of the head) is used to read and write induction is the process where a moving magnetic field can create a current in a wire the platters spin

For reading, the magetized coating passing under the induction coil which can create (or not create) a current. The current / no current detection from the coil is translated into the 0's and 1's. Note that usually it is the change from current to no current that is used as the "bit".

"The man who does not read good books has no advantage over the man who can't read them." - Samuel Clemens (a.k.a, Mark Twain)

For writing, a current is passed through the coil, and since an electrical field always has a surrounding magnetic field, this is used to set or magnetize the coating. This same basic idea is used for cassette tapes, VHS tapes (Betamax and 8-tracks too!) only the physical storage media is a sequential and a plastic film substrate.

Induction coil does not make contact with the platter, if it does, this would be a "crash", and is very bad. The air currents created by the platter spinning allow the r/w head to "float" just above the platter for the most part, data can be accessed randomly rather than sequentially (a big advantage to tapes) the head resides on an arm that moves it radially, so with the disk spinning can get to any point of the disk quickly To increase capacities, multiple platters are used stack on top of each other Each platter has its own R/W head, though most typically all are attached to the same arm control mechanism.

track = yellow, sector = blue

sectors are fized sized (usually 512 bytes)

cylinders are similar to tracks, across multiple platters

• preamble delimits sectors
• Extremely precise devices, platters/heads sealed since small dust particles would ruin the drive
• ECC - error correcting code - after sector (similar to Hamming Code but allows multiple errors to be fixed

• Radial and circumferencial densities different
• low-level formatting - where are the tracks, the sectors
• high-level formatting - OS dependent for managing the data (mapping filenames to disk locations, managing free space, etc.)
• unformatted capacity includes preambles, ECCs and gaps , so formatted capacity around 15% less
• four things dictate the disk drives capacity:
  1. The number of cylinders that the disk contains;
  2. The number of tracks per cylinder (same as the number of heads);
  3. The number of sectors per track; and,
  4. The size of each sector (in bytes).

This is the last disk drive I bought: model: WD 800BB - 00DAA1 number of disks = 3 number of surfaces = 6 number of heads = 6 bytes per sector = 512 user sector per drive = 156,301,488 capacity = 512 * 156,301,488 = 80,026,361,856 ~ 80GB

"There are only two truly infinite things, the universe and stupidity. And I am unsure about the universe." - Albert Einstein

• modern drives have differing numbers of sectors per track depending on how far from center it is.

How a disk retrieves/writes data: move arm so head is on right track/cylinder wait for data to rotate under the R/W head read/write data seek - moving arm to correct radial position SPECS Average Read Seek 8.9 ms (average) Track-to-Track Seek 4 2.0 ms (average) Full Stroke Read Seek 4 21 ms (average) rotational latency - waiting for data to spin under head (depends on disk's rotational speed) SPECS 7200 RPM = 120 RPS (revolutions per second) = 0.0083 SPR (seconds per revolution) = 8.3 milliseconds On average disk has to sping half way round, do the ave rotational latency is about 4.2 milliseconds transfer times - read/write the data, depends on linear density and rotational speed SPECS Up to 100 MB/s burst transfer rates 195,312 sectors/sec 0.00000512 secs per sector 5.12 microsec Notice that seek times and latency times are 3 orders of magnitdue slower than the transfer times. maximumn burst rate versus maximum sustained rate - burts rate does not include the time spent over the ECCs, preambles and gaps

• a cpu on the drive that handles read/write requests and controllers the disk operations
• modern drives have caches to speed things up
• exploiting same idea that data is not accessed randomly.

Floppy disk drives were originally introduced commercially as a read-only device to hold microcode and diagnostics for large IBM mainframe computer systems in the early 1970s. Used to be the way softare was distributed: no DVDs no CDs no ZIP drives no Internet Layout similar to hard drives, but: Plastic film (floppy) disk rather than rigid aluminun platter R/W head touches surface slower less durable extra "spin up" times

• invention (1967) (IBM)
• large 8", 100 KB (IBM) (1971)
• large 8", 150 KB (1973)
• 5.25", one sided, low density, 90K (1976) (Shugart/Dysan)
• 5.25", one sided, double density, 180 KB
• 5.25", double sided, double density, 360 KB
• 5.25", two sided, high density density, 1.2MB
• 3.5". double sided, double density, 720KB (1980) (SONY)
• many competing floppy variations (early 80's)
• 3.5", double sided, double density, 1MB (1980) (SONY)
• 3.5", double sided, high density, 1.44 MB

• in PCs used to be on a separate board
• I/O done by writing head/cyl/sect to registers and calling BIOS handling routine (special ISA level instruction)
• This special instruction allows the BIOS to worry about the hardware and provides a standard view for interfacing with disk.
• IDE - integrated drive electronics (mid-80's) has controller integrated with the drives
• BIOS allowed 24 bits for the specification of the drive's geometry, broken up as follows:
• But IDE had its own limits, which combined like this:

• NOTE: 504 MB = 528,482,304 million bytes
• I/O calling convention retained for backward compatibility
• drives with wrong geometry (e.g., > 1024 cylinders) would lie about geometry and do the mapping themselves
• drives above 528 MB required new BIOS allowing for translating, though still required disk controller to "lie" about geometry.
• there were many weird limits along the way, each requiring some tweak. (504MB, 4GB, 8GB)
• an enhanced IDE (EIDE) was also needed with new addressing mode linear block addressing (LBA)

• physically similar to IDE disks
• different interface and different bus
• faster
• transfer rates for IDE's catching up, but SCSI still seems faster, especially for concurrent I/O

• usually used in higher end systems and servers
• SCSI controller can chain up to 7 devices

• even the fastest disks are still orders of magnitude slower than memory and processors
• using parallelism at the disk level is the next logical way to get speedup
• RAID - redundant array of inexpensive disks
• RAID addresses speed and reliability
• various RAID schemes ("levels") each which is geared toward some level of speed and/or reliability
• levels is a misnomer as there is no real hierarchy involved

• basic idea is to have a RAID controller card present itself to the processor/OS as just another disk drive.
• internally the controller deals with manging and mapping the data to a series of disks
• specific disk organization determines how fast and/or reliable the system will be.
• Most RAID systems are SCSI based since it is faster an allows more disks to be connected

• group sectors into stripes
• write stripes in round-robin fashion
• parallel I/O

(+) if large data requests (multiple stripes) (-) if small data requests (less than 1 stripe) (-) reliability (1 of 4 failing is more likely than just 1 failing)

• like Level 0, but duplicate each stripe
• write takes same time (write twice, but in parallel)

(+) reads twice as fast (can use either copy) (+) immune to disk single disk failures (-) still bad for small requests

• data split into bits, nibbles or bytes and written across disks
• use Hamming-like code to do error detetion/correction
• Example of splitting bytes into nibbles and using 3 extra bits for Hamming Code:

(+) immune to single disk failures (+) less overhead (Level 1 is 100% overhead) (+) speedup for large and small requests (-) need rotational synchronization (-) more complex/faster controller needed

• split into bits, nibbles or bytes like level 2
• only use parity bit
• Example using nibbles

(+) less overhead (one bit not three) (+) can do error correction even with just parity bit, since we will know which drive failed. (-) need rotational synchronization (-) more complex/faster controller needed

• uses stripes like Levels 0 and 1
• uses parity like level 3, but on a stripe-by-stripe basis
• dedicated drive for stripe parities

(+) do not need synchronized drives (+) immune to single disk crashes (+) parallel I/O (-) slow for small write since all drives must be read to recompute parity (-) parity drive used in all operations, so becomes a bottleneck

• like Level 4, but distributes parity across drives to "load balance" them.

• reconstructing failed drive is more complex

original phyical design for audio Writing laser burns 0.8 micron holes in a glass disk use this "master" to create a mold polycarbonate resin injected into molds coat with aluminum and protective laquer result is a disc with small "pits" and "lands" Reading read head with a laser can then read differences in height to distinguish 0's and 1's (using destructive interference) actually use the transitions from lands to pit to hold information, since this is more reliable than trying to tell absolute depths. data stored as a single continuous spiral groove for audio, data must stream at uniform rate, so disk must slow down as it goes from inside to outside.

• 1983 - Philips and Sony develop the CD-ROM, as an extension of audio CD technology.
• used same phyical and optical characteristics, but defined data format and improved error correction
• book has some gorey details of this format

• CD-ROM standard was the low-level format of a CD-ROM
• also need high-level format (filesystem) to be standard: ISO9660
• three levels Level 1 - DOS-ish 8.3 filenames, case-insensitive, contiguous files, limited directory nesting Level 2 - names up to 32 char Level 3 - non-contiguous files Rock Ridge Extensions - unix-like felxibility

• capacity 650 MB
• single speed 175.2 KB/sec
• Even 56X CD-ROM drive is only 8.2 MB/sec, while even low-end SCSI can get 10 MB/sec

uses dye to alter brightness of relfected laser light chemical properties of dye: transparent normally, turning dark when heated to "burn" higher powered laser is used to heat up areas needed to encoded the data see book for more details Also came up with CD-ROM XA which allows a CD-R to be written incrementally writing must be done in a single continuous operation, if data cannot be fed fast enough, CD-R is ruined

yet another technology uses a material that has two states, with different reflectiveity depending on intensity of laser light it will read the data or force it to go through a state transition thus changing its reflectivity more expensive than CD-Rs

originally for audio/video same idea as CDs, but can get higher densities with improved technology different laser, so need two if CD's are to be read capacity 4.7 GB single speed 1.4 MB/sec still evolving DVD-R, DVD-RW, and many other formats, e.g., DVD-VCD

• 1994 - DEC. Iomega Corp. introduces its Zip drive and Zip disks, floppy disk sized removable storage in sizes of 25MB or 100MB.
• big, fast, removable, rewritable
• these looked to be the future as floppy disks had been overwhelmed by the size of software and storage requirements of the average user
• a few competitors entered the market
• CD-R and the plummenting cost per megabyte of hard drives have kept these from taking off too big. e.g., I mirror all my data on another hard drive