CSE451 Notes for 5/21/03
Devon Kim
Silas Brazil
Topic: Secondary Storage and Disk Structure Introduction
Point: Secondary Storage is the larger, slower chunk of the memory hierarchy,
and can also be virtualized.
**** VIEWING NOTES AND SUGGESTIONS ******
This page contains ascii graphics - please use a fixed width font, like Courier
New, to view it. For the best viewing (or printing) experience, please view
using a classic console font (ie. try using Lynx / Links or your favorite
Unix text editor).
ASCII art was found through Google search terms "ASCII" "art" and "download"
Original art was modified (heavily) to suit needs.
ASCII Generator was used in the making
of the graphs as well as the ubiquitous Microsoft Paint (XP).
-------------------------------------------------------------------------------------------------------------------------------------------------
--Quick review of class organization:
part 1: Processor Virtualization
part 2: Memory Virtualization
part 3: Storage Virtualization
-------------------------------------------------------------------------------------------------------------------------------------------------
--Secondary Storage Definition and Facts
Secondary storage refers to all mass storage persistent media (anything other
than volatile main memory): CDs,magnetic disk, Flash.
Tertiary storage includes magnetic tape.
Every year, secondary storage capacity doubles for the same price (now about
$0.25/Mb).
All secondary storage media share the same properties: they are persistent, and
their price is much less than main
mem, but access time is longer.
Graph of storage density (Mbits per square inch) from 1960 to present: the past
40 years have improved density
by a factor of 5 million
M; M
M; ME
M; M
M; MM
M; ZM
M; MMM
MB/in^2 M; MM
M; MMM.
M; MMMM
( log ) M; HMMM
( scale) M; JMMMMJ
M; NMMMMH
M; .MMMMMM;
M; .MMMMMN
M; 2MMMMM
M; mMMMK
M; MMML
M; MMM
M; MM2
M; YM
M; MM
M ML
M MU
MLM
MM
MMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMM
1960 year 2000
^
|
| Overall improvement: 5 MILLION X
|
(professor's sidenote: machines back then used HUGE, unwieldy, clunky disks)
-------------------------------------------------------------------------------------------------------------------------------------------------
--Memory Hierarchy:
Memory is distributed from the very small, fast volumes close to the CPU to the
large, slow volumes far from the CPU.
Smaller memory is more expensive, has a smaller address space, is volatile, and
is quickly accessible. Larger memory
is the opposite in every way. Technological improvements naturally bring faster
memory closer to the CPU in order to
improve performance by exploiting locality of reference. Price also determines
how much of the faster memory can be practically available, and so the more expensive
resources are placed close to CPU, where they will be used more often.
The memory hierarchy, with average access times:
Memory hierarchy:
bigger slower
Access time |===========| \ | |
| | | \ | |
1 cycle | registers | compiler manages V V
| | | /
| |===========| /
| |
| |=================| \
| | | \
< 5 cycles | L1 cache | \
| | | architecture manages
| |=================| |
| | |
| . |======================| |
| . | | /
| . | L2 cache | /
| . | | /
| . |======================| /
| . |
| . |=============================| \
| . | | \
~20 cycles | physical memory | \
| . | | \
| . |=============================| |
| . | |
| . |===================================| operating system
| . | | manages
| . | secondary storage | |
~1 M cycles |(hard disks, optical storage, etc)| |
| . | | |
| . |===================================| /
| . | /
| . |===========================================| /
| | tertiary storage | /
~minutes | (tape storage - God save us) |
to days |===========================================|
-------------------------------------------------------------------------------------------------------------------------------------------------
--Magnetic Disk Structure
The hard drive is arranged as a stack of magnetic platters skewered on a spindle.
Each platter is double-sided and is
read/written by a magnetic head attached to a common armature. Each side of each
platter is concentrically divided into
tracks, and each track is divided radially into sectors. Each sector (aka 'block')
contains some amount of bits.
The average hard drive contains 8-16 platters, a few thousand tracks per platter,
and a variable (depending on radius)
number of sectors per track.
A picture of the hard drive structure:
armature w/ read/write heads over and under each platter
_.-----._
|>=====| -. platter 0
|| -_ _-
|| ~-----~
|>=====| <-----------------------underneath r/w head
||
|| _.-----._
|>=====| -. platter 1
|| -_ _-
|| ~-----~
|>=====|
||
|| _.-----._
|>=====| -. platter 2
|| -_ _-
|| ~-----~
|>=====|
||
|| _.-----._
|>=====| -. platter 3
|| -_ _-
|| ~-----~
...
-------------------------------------------------------------------------------------------------------------------------------------------------
--Disk Address Virtualization
Information in a hard drive is internally address by sector, giving rise to the
four-tuple internal address format:
{ side . platter# . track# . sector# }
The differences in hard drive internals between manufacturers, changes across
technological improvements, and the general complexity of the internal addressing
scheme motivate disk-addressing virtualization.
This virtualization scheme is simply achieved by linearly addressing sectors, and
allowing the addressor to ignore all
internal structure(other than the total number of sectors). Thus the OS that
requests a block of data from the disk may
do so simply by referring to a unique block number.
(Note: if you're using a font with funny dimensions, the following
will look like an ellipse. Real disks are circles, not ellipses)
A disk platter:
,,ggddY""""Ybbgg,,
,agd""' | `""bg,
,gdP" sector | sector "Ybg,
,dP"\ 11 | 10 / "Yb,
,dP" \ _,,ddP"""Ybb,,_ / "Yb,
,8" \ ,dP"' `"Yb, / "8,
,8' sector ,d" \ sector 3 / "b, sector`8,
,8' 12 d" \ / "b 9 `8,
d' d'sector \,gPPRg,/sector `b `b
8 8 4 dP' `Yb 2 8 8
8---------8--------8) O (8--------8---------8
8 8 sector Yb dP sector 8 8
8 Y, 5 /"8ggg8"\ 1 ,P 8
Y, sector Ya / \ aP sector ,P
`8, 13 "Ya / (track 0) \ aP" 7 ,8'
`8, / "Yb,_ sector 0_,dP"\ ,8'
`8a / `""YbbgggddP""' \ a8'
`Yba / sector | sector \ adP'
"Yba/ 14 (track 1) 6 \ adY"
`"Yba, | ,adP"'
`"Y8ba, | ,ad8P"'
``""YYbaaadPP""''
-------------------------------------------------------------------------------------------------------------------------------------------------
--Disk Access Time Definitions
* "Internal Disk Transfer Time", usually about 1ms
The amount of time needed to collect a block (sector) of data *that is already
under the read/write head*. (This could be multiple sectors if the request was
large, but to be useful as a metric, Internal Disk Transfer Time would need to
be expressed as time per constant amount of data.)
* "Rotational Latency" = on average 0.5 * 1/rotation_speed, usually about
5-10msec
An issue when the requested sector is not under the read/write head, but the
read/write head is on the correct track.
num_sectors is the number of sectors that must be passed by to read the requested
sector that is not under the head.
* "Seek Time", usually about 10-30ms
The time required to move the read/write head when it it not on the requested
track.
-------------------------------------------------------------------------------------------------------------------------------------------------
--General Design Issues
* Power Consumption:
More power to the motors increases disk access speed, but this is a limitation
in portable devices.
Several methods of motor powering:
-Outlet: roughly 7200 - 15K rpm, with a seek time less than 10ms
-Internal Battery: 3600 rpm, average seek time of 50ms
*Density:
Each bit in a sector is measured by a rectangular width and height. The mechanical
engineers reduce the height by improving the precision of the read/write heads
(narrowing the track width), while physicists work on reducing the width (by improving
the sensitivity of the read/write heads).
,,ggddY""""Ybbgg,,
,agd""' ^ `""bg,
,gdP" \ Height / "Ybg, better through
,dP" \ v ----/------------ higher granularity of
,dP" \_,,ddP"""Ybb,,_/ "Yb, head movements
,8" ,dP"' ^ `"Yb, "8,
,8' ,d" <----Width----> ------------------- better
,8' d" "b `8, with physics
d' d' ,gPPRg, `b `b & bigger sectors
8 8 dP' `Yb 8 8
8 8 8) O (8 8 8
8 8 Yb dP 8 8
8 Y, "8ggg8" ,P 8
Y, Ya aP ,P
`8, "Ya aP" ,8'
`8, "Yb,_ _,dP" ,8'
`8a `""YbbgggddP""' a8'
`Yba adP'
"Yba adY"
`"Yba, ,adP"'
`"Y8ba, ,ad8P"'
``""YYbaaadPP""''
-------------------------------------------------------------------------------------------------------------------------------------------------
--For Next Lecture:
Bigger sectors allow for better error correction...