Lec02 secondaryStorageDevices
Lec02 secondaryStorageDevices
Lec02 secondaryStorageDevices
CENG 351
2. Serial Devices
CENG 351
CENG 351
Storage Hierarchy
Primary storage : random access memory (RAM) typical capacity a number of GB cost per MB $2-3.00 typical access time 5ns to 60ns Secondary storage: magnetic disk/ optical devices/ tape systems typical capacity a number of 100GB for fixed media; g for removable cost per MB $0.01 for fixed media, more for removable typical access time 8ms to 12ms for fixed media, larger for removable
CENG 351 4
Units of Measurement
Spatial units: o byte: 8 bits o kilobyte (KB): 1024 or 210 bytes o megabyte (MB): 1024 kilobytes or 220 bytes o gigabyte (GB): 1024 megabytes or 230 bytes Time units: o nanosecond (ns) one- billionth (10-9 ) of a second o microsecond ( Qs) one- millionth (10-6 ) of a second o millisecond (ms) one- thousandth (10-3 ) of a second Primary versus Secondary Storage Primary storage costs several hundred times as much per unit as secondary storage, but has access times that are 250,000 to 1,000,000 times faster than secondary storage.
CENG 351 5
Memory Hierarchy
At the primary storage level, the memory hierarchy includes, at the most expensive end cache memory, which is a static RAM (Random Access Memory). The next level of primary storage is DRAM (Dynamic RAM), The advantage of DRAM is its low cost, lower speed compared with static RAM. Programs normally reside and execute in DRAM. Now that personal computers and workstations have 10s of gigabytes of data in DRA, in some cases, entire databases can be kept in the main memory (with a backup copy on magnetic disk), leading to main memory databases.
CENG 351
CENG 351
Magnetic Disks
Bits of data (0s and 1s) are stored on circular magnetic platters called disks. A disk rotates rapidly (& never stops). A disk head reads and writes bits of data as they pass under the head. Often, several platters are organized into a disk pack (or disk drive).
CENG 351 8
A Disk Drive
surfaces
Looking at a surface
tracks
sector
Surface of disk showing tracks and sectors
CENG 351 10
Organization of Disks
Disk contains concentric tracks. Tracks are divided into sectors A sector is the smallest addressable unit in a disk.
CENG 351
11
Components of a Disk
Disk head Spindle Tracks
The arm assembly is moved in or out to position a head on a desired track. Tracks under heads make a cylinder (imaginary!). Only one head reads/writes at any one time.
Arm movement
Platters
Arm assembly
CENG 351
12
Disk Controller
Disk controllers: typically embedded in the disk drive, which acts as an interface between the CPU and the disk hardware.
The controller has an internal cache (typically a number of MBs) that it uses to buffer data for read/write requests.
CENG 351 13
Accessing Data
When a program reads a byte from the disk, the operating system locates the surface, track and sector containing that byte, and reads the entire sector into a special area in main memory called buffer. The bottleneck of a disk access is moving the read/write arm.
So it makes sense to store a file in tracks that are below/above each other on different surfaces, rather than in several tracks on the same surface.
CENG 351 14
Cylinders
A cylinder is the set of tracks at a given radius of a disk pack.
i.e. a cylinder is the set of tracks that can be accessed without moving the disk arm.
All the information on a cylinder can be accessed without moving the read/write arm.
CENG 351
15
Cylinders
CENG 351
16
Estimating Capacities
Track capacity = # of sectors/track * bytes/sector Cylinder capacity = # of tracks/cylinder * track capacity Drive capacity = # of cylinders * cylinder capacity Number of cylinders = # of tracks in a surface
CENG 351
17
Exercise
Store a file of 20000 records on a disk with the following characteristics:
# of bytes per sector = 512 # of sectors per track = 40 # of tracks per cylinder = 11 # of cylinders = 1331
Q1. How many cylinders does the file require if each data record requires 256 bytes? Q2. What is the total capacity of the disk?
CENG 351 18
Exercise
Suppose we want to read consecutively the sectors of a track in order: sectors 1, 2,11. Suppose two consecutive sectors cannot be read in non-interleaving case. How many revolutions to read the disk?
a) Without interleaving b) With 3:1 interleaving Note: nowadays most disk controllers are fast enough so interleaving is not common...
CENG 351 20
Clusters
Usually File manager, under the operating system, maintains the logical view of a file. File manager views the file as a series of clusters, each of a number of sectors. The clusters are ordered by their logical order. Files can be seen in the form of logical sectors or blocks, which needs to be mapped to physical clusters. File manager uses a file allocation table (FAT) to map logical sectors of the file to the physical clusters.
CENG 351 21
Extents
If there is a lot of room on a disk, it may be possible to make a file consist entirely of contiguous clusters. Then we say that the file is one extent. (very good for sequential processing) If there isnt enough contiguous space available to contain an entire file, the file is divided into two or more noncontiguous parts. Each part is a separate extent.
CENG 351 22
Internal Fragmentation
Internal fragmentation: loss of space within a sector or a cluster.
1) Due to records not fitting exactly in a sector: e.g. Sector size is 512 and record size is 300 bytes. Either store one record per sector, or allow records span sectors 2) Due to the use of clusters: If the file size is not a multiple of the cluster size, then the last cluster will be partially used.
CENG 351 23
CENG 351
24
Non-data Overhead
Both blocks and sectors require non-data overhead (written during formatting) On sector addressable disks, this information involves sector address, track address, and condition (usable/defective). Also pre-formatting involves placing gaps and synchronization marks between the sectors. On block-organized disk, where a block may be of any size, more information is needed and the programmer should be aware of some of this information to utilize it for better efficiency
CENG 351 26
Exercise
Consider a block-addressable disk with the following characteristics:
Size of track 20,000 bytes. Nondata overhead per block = 300 bytes. Record size = 100 byte.
Q) How many records can be stored per track if blocking factor is 10 or 60? a) 10 (20000/1300*10=150) b) 60 (20000/6300*60=180)
CENG 351 27
Time Component
Seek Time Rotational delay (or latency) Transfer time
Action
Time to move the read/write arm to the correct cylinder Time it takes for the disk to rotate so that the desired sector is under the read/write head Once the read/write head is positioned over the data, this is the time it takes for transferring data
CENG 351 28
Seek time
Seek time is the time required to move the arm to the correct cylinder. Largest in cost. Typically:
5 ms (miliseconds) to move from one track to the next (track-to-track) 50 ms maximum (from inside track to outside track) 30 ms average (from one random track to another random track)
CENG 351
29
If the starting positions for each access are random, it turns out that the average seek traverses one third of the total number of cylinders.
Why? There are more ways to travel short distance than to travel long distance
Manufacturers specifications for disk drives often list this figure as the average seek time for the drives. Most hard disks today have s under 10 ms, and high-performance disks have s as low as 7.5 ms. 30 CENG 351
Then the time for the head to move n tracks is: Seek(n)= Hs+ Ht*n
CENG 351
31
Note:
Min latency = 0 Max latency = Time for one disk revolution Average latency (r) = (min + max) / 2 = max / 2 = time for disk revolution Typically 6 4 ms, at average
CENG 351 32
Rotational Latency-2
Given the following:
R = the rotational speed of the spindle (in rotations per second) U = the number of radians through which the track must rotate then the rotational latency U radians is:
Latency= (U/24)*(1000/R), in ms
CENG 351
33
Transfer Time-1
Transfer time is the time for the read/write head to pass over a block. The transfer time is given by the formula:
Transfer time = number of sectors --------------------------------- x rotation time track capacity in number of sectors
e.g. if there are St sectors per track, the time to transfer one sector would be 1/ St of a revolution.
CENG 351 34
Transfer Time-2
The transfer time depends only on the speed at which the spindle rotates, and the number of sectors that must be read. Given:
St = the total number of sectors per track the transfer time for n contiguous sectors on the same track is:
MB1
Exercise
Given the following disk:
20 surfaces 800 tracks/surface 25 sectors/track 512 bytes/sector 3600 rpm (revolutions per minute) 7 ms track-to-track seek time 28 ms avg. seek time 50 ms max seek time.
Find:
a) Average latency b) Disk capacity c) Time to read the entire disk, one cylinder at a time
CENG 351 36
Slide 36 MB1 Average latency=(60*1000/3600)/2=10 ms Disk capacity=20*800*25*512=20MB Time to read the entire disk=20*800*7=112 sec
Muslim Bozyigit, 9/28/2006
MB2
Exercise
Disk characteristics:
Average seek time = 8 msec. Average rotational delay = 3 msec Maximum rotational delay = 6 msec. Spindle speed = 10,000 rpm Sectors per track = 170 Sector size = 512 bytes
Sequential Reading
Given the following disk:
Avg. Seek time s = 16 ms Avg. Rot. Latency r = 8.3 ms Block transfer time = 8.4 ms
MB3
a) Calculate the time to read 10 sequential blocks, on the same track. b) Calculate the time to read 10 sequential cylinders, if there are 200 cylinders, and 20 surfaces.
CENG 351 38
Random Reading
Given the same disk, a) Calculate the time to read 100 blocks randomly b) Calculate the time to read 100 blocks sequentially.
CENG 351
39
Exercise
Given a file of 30000 records, 1600 bytes each, and block size 2400 bytes, how does record placement affect sequential reading time, in the following cases? Discuss.
i) Empty space in blocks-internal fragmentation. ii) Records overlap block boundaries.
CENG 351
42
Exercise
Specifications of a disk drive:
Min seek time, track-to-track = 6ms. Average seek time = 18ms Rotational delay = 8.3ms Transfer time or byte transfer rate=16.7 ms/track or 1229 bytes/ms Bytes per sector = 512 Sectors per track = 40 Tracks per cylinder = 12 (number of surfaces) Tracks per surface = 1331 Non Interleaving Cluster size= 8 sectors Smallest extent size = 5 clusters
CENG 351
44
Characteristics
No direct access, but very fast sequential access. Resistant to different environmental conditions. Easy to transport, store, cheaper than disk. Before it was widely used to store application data; nowadays, its mostly used for backups or archives.
CENG 351 45
MT Characteristics-2
A sequence of bits are stored on magnetic tape. For storage, the tape is wound on a reel. To access the data, the tape is unwound from one reel to another. As the tape passes the head, bits of data are read from or written onto the tape.
CENG 351 46
Reel 1
Reel 2
Tracks
Typically data on tape is stored in 9 separate bit streams, or tracks. Each track is a sequence of bits. Recording density = # of bits per inch (bpi). Typically 800 or 1600 bpi. 30000 bpi on some recent devices.
CENG 351
48
MT recording in detail
8 bits = 1 byte
0 1 1 0 1 1 0 1 0
0 1 1 0 1 1 0 1 0
0 1 1 0 1 1 0 1 0
0 1 1 0 1 1 0 1 0
parity bit
CENG 351 49
Tape Organization
logical record
2400
Data blocks
EOT marker
How many records can we write on the tape? (ignoring BOT and EOT markers and the header block for simplicity)
CENG 351 52
CENG 351
53
Organization of data
Reading a CD is done by shining a laser at the disc and detecting changing reflections patterns.
1 = change in height (land to pit or pit to land) 0 = a fixed amount of time between 1s LAND PIT LAND PIT
LAND
Note : we cannot have two 1s in a row! => uses Eight to Fourteen Modulation (EFM) encoding table. Usually, a pattern of 8 bits is translated to/from a pattern of 14 pits and lands.
CENG 351 55
CD-ROM
y While the speed of CD-ROM readers is relatively higher, such as 24X(24 times CD audio speed), the speed of writing is much slower, as low as 2X.
Note that the speed of the audio is about 150KB per second.
yThe DVD (Digital Video Disc or Digital Versatile Disc) technology is based on CD technology with increased storage density. The DVD technology allows two-side medium, with a storage capacity of up to 10GB.
CENG 351 56
CD-ROM
Because of the heritage from CD audio, the data is stored as a single spiral track on the CD-ROM, contrary to magnetic hard disks discrete track concept. Thus, the rotation speed is controlled by CLV-Constant Linear velocity. The rotational speed at the center is highest, slowing down towards the outer edge. Because, the recording density is the same every where. Note that with CLV, the linear speed of the spiral passing under the R/W head remains constant. CLV is the culprit for the poor seek time in CD-ROMs The advantage of CLV is that the disk is utilized at its best capacity, as the recording density is the same every where.
CENG 351 57
CD-ROM
Note that: Since 0's are represented by the length of time between transitions, we must travel at constant linear velocity (CLV)on the tracks. Sectors are organized along a spiral Sectors have same linear length Advantage: takes advantage of all storage space available. Disadvantage: has to change rotational speed when seeking (slower towards the outside)
CENG 351 58
CD-ROM
Question: Why does it take only 70 minutes of playing time in an CD audio. y Ans.: If the sound frequency is 20 kilohertz, we need twice as much frequency for sampling speed to reconstruct the sound wave. Each sample may take up to 2 bytes.
-An accepted standard allows a sampling speed of 44100 times per second, which requires 88200 bytes for 2 bytes per sample. For stereo, this becomes 176400 bytes per second. If -If the capacity is about 600MB, you can compute number of minutes required
CENG 351 59
Addressing
1 second of play time is divided up into 75 sectors. Each sector holds 2KB 60 min CD: 60min * 60 sec/min * 75 sectors/sec = 270,000 sectors = 540,000 KB ~ 540 MB A sector is addressed by: Minute:Second:Sector e.g. 16:22:34
CENG 351 60
If implement UNIX file system on CD-ROM, it will be a catastrophe! The seek time per access is from 500 msec to 1 sec.
CENG 351 61
For a small file structure file, the entire directory structure can be kept in the memory all the time, which allows method to work.
CENG 351 62
CENG 351
64
Representation of individual files on CD-ROM B+ Tree type data structures are appropriate for organizing the files on CD-ROMs. Build once read many times allows attempting to achieve100% utilization of blocks or buckets. Packing the internal nodes so that all of them can be maintained in the memory during the data fetches is important. Secondary indexes can be formed so that the records are pined to the indexes on a CD-ROM, as the file will never be reorganized
CENG 351 65
Representation of individual files on CD-ROM This may force the files on the source disks and their copies on the CD-ROM to be differently organized, because of the efficiency concerns. It is possible to use hashing on the CD-ROM, except that the overflow should either not exist or minimized. This becomes possible when the addressing space is kept large. Remember that the files to be put on a CD-ROM are final, so the hashing function can be chosen to perform the best, i.e. with no collisions.
CENG 351 66
CENG 351
67
A journey of a byte
Suppose in our program we wrote:
outfile << c;
This causes a call to the file manager (a part of O.S. responsible for I/O operations) The O/S (File manager) makes sure that the byte is written to the disk. Pieces of software/hardware involved in I/O:
Application Program Operating System/ file manager I/O Processor Disk Controller
CENG 351 68
Application program
Requests the I/O operation
I/O Processor
a separate chip; runs independently of CPU Find a time when drive is available to receive data and put dat in proper format for the disk Sends data to disk controller
Disk controller
A separate chip; instructs the drive to move R/W head Sends the byte to th surface when the proper sector comes under R/W head.
CENG 351 69
Buffer Management
Buffering means working with large chunks of data in main memory so the number of accesses to secondary storage is reduced. Today, well discuss the System I/O buffers. These are beyond the control of application programs and are manipulated by the O.S. Note that the application program may implement its own buffer i.e. a place in memory (variable, object) that accumulates large chunks of data to be later written to disk as a chunk. Read Section 4.2 for using classes to manipulate program buffers.
CENG 351 70
Buffer Bottlenecks
Consider the following program segment:
while (1) { infile >> ch; if (infile.fail()) break; outfile << ch; }
Buffering Strategies
Double Buffering: Two buffers can be used to allow processing and I/O to overlap.
Suppose that a program is only writing to a disk. CPU wants to fill a buffer at the same time that I/O is being performed. If two buffers are used and I/O-CPU overlapping is permitted, CPU can be filling one buffer while the other buffer is being transmitted to disk. When both tasks are finished, the roles of the buffers can be exchanged.