`
`wavarehivcerg
`415.561.6736?
`415340—0391 e—fax
`
`Intemet _-\rchive
`300 I'lmston Avenue
`
`San Francisco, CA 94] IS
`
`AFFIDAVIT 0F CHRISTOPHER BUTLER
`
`I am the Office Manager at the Internet Archive, located in San Francisco,
`1.
`California. I make this declaration of my own personal knowledge.
`2. The'Internet Archive is a website that provides access to a digital library of
`Internet sites and other cultural artifacts in digital form. Like a paper library, we provide
`free access to researchers, historians, scholars, and the general public. The Internet
`Archive has partnered with and receives support from various institutions, including the
`Library of Congress.
`3. The Internet Archive has created a service known as the Waybaek Machine. The
`Wayback Machine makes it possible to surf more than 450 billion pages stored in the
`Internet Archive's web archive. Visitors to the Wayback Machine can search archives
`by URL (i.e., a website address). If archived records for a URL are available, the visitor
`will be presented with a list ofavailable dates. The visitor may select one of those
`dates, and then begin surfing on an archived version ofthe Web. The links on the
`archived files, when served by the Wayback Machine, point to other archived files
`(whether HTML pages or images). Ifa visitor clicks on a link on an archived page, the.
`Wayback Machine will serve the archived file with the closest available date to the page
`upon which the link appeared and was clicked.
`4. The archived data made viewable and browseable by the Wayback Machine is
`compiled using software programs known as crawlers, which surfthe Web and
`automatically store copies of web files, preserving these files as they exist at the point of
`time of capture.
`5. The Internet Archive assigns a URL on its site to the archived files in the format
`http:fa’webarchive.org/webi’[Year in yyyy][Month in mm][Day in dd][Time code in
`hh:mm:ss]f[Archived URL]. Thus, the Internet Archive URL
`http:i’fwebarchiveorg/webi’l99?0l2604582Br‘http:fiwwwarchiveorgf would be the
`URL for the record of the Internet Archive home page HTML file
`(httpflwwwarchiveorgfl archived on January 26, 199? at 4:58 am. and 28 seconds
`(199WOH26 at 04:58:23). A web browser may be set such that a printout from it will
`display the URI. ofa web page in the printout’s footer. The date assigned by the Internet
`Archive applies to the HTML file but not to image files linked therein. Thus images that
`appear on a page may not have been archived on the same date as the HTML file.
`Likewise, ifa website is designed with "frames," the date assigned by the Internet
`Archive applies to the frameset as a whole, and not the individual pages within each
`frame.
`
`6. Attached hereto as Exhibit A are true and accurate copies of printouts ofthe
`Internet Archive's records of the HTML files or .ps files for the URLs and the dates
`specified in the footer of the printout (HTML) or attached coversheet (.ps).
`7. I declare under penalty of perjury that the foregoing is true and correct.
`
`DATE:
`
`i
`
`IS 7")
`
`/_ . ___,_
`Christopher Butler
`
`Qualcomm Incorporated
`Exhibit 1033
`
`Page 1 of 14
`
`Qualcomm Incorporated
`Exhibit 1033
`Page 1 of 14
`
`
`
`
`Exhibit A
`
`Exhibit A
`
`Page 2 of 14
`
`Page 2 of 14
`
`
`
`http://web.archive.org/web/19970715044633/http://www.cis.upenn.edu/~udani/papers/surv
`ey.ps
`
`
`Page 3 of 14
`
`
`
`Power Management in Mobile Computing*
`
`Jonathan Smith
`Sanjay Udani
`Distributed Systems Laboratory
`Department of Computer Information Science
`University of Pennsylvania
`{udam', jms} @dsl. 02's. upenn. edu
`
`August 1996
`
`Abstract
`
`Rapid advances in technology have resulted in laptop (mobile) computers with performance and
`features comparable to desktop (stationary) machines. Advances in rechargeable battery technology
`have failed to keep pace, decreasing the usefulness of mobile computers and portable wireless devices.
`Several methods of power management can be used to prolong the battery life of a mobile computer.
`We provide a detailed analysis of power consumption typically encountered in a networked laptop com-
`puter and the power management methods currently used. We also outline some novel proposed power
`management methods.
`
`1
`
`Introduction
`
`Laptop computers have often served as portable word processors or game machines. Such machines were
`generally two or more generations behind desktop computers in terms of processing power, features and
`performance. Limitations in display and miniaturization technology prevented laptops from being able to
`compete with desktops as “real” (i.e. full featured) computers.
`Recent advances in technology have dramatically improved laptop performance and it is increasingly
`common to see software development being done on a laptop. Laptops with a 133 MHZ Pentium processor,
`1.2 Gigabyte hard disk, modular 6X CD ROM drive and 12.1 inch SVGA display are available in mid—1996,
`albeit at a price premium over comparable desktops. A survey in Computerworld [7] predicts that the
`number of workers using portable computers will expand from about one in five today to about one in three
`by the year 2000, and that 80% of portable users will use their portables as their primary machines, up
`from the current 30%. This optimistic view is heavily dependent on laptops being able to overcome some
`key drawbacks. In addition to a price premium, laptops have another significant disadvantage compared to
`desktopsilimited battery life.
`
`1.1 Background
`
`liquid crystal display (LCD),
`The major components of a typical laptop are the microprocessor (CPU),
`hard disk, system memory (DRAM), keyboard/mouse, CD ROM drive, floppy drive, I/O subsystem, audio
`subsystem and in the case of a mobile computer, a wireless network card. There are other components, but
`these are significant consumers of power. The CPU /motherboard of a laptop poses several design problems
`not found in a desktop.
`In addition to the power it consumes, there are also extreme thermal dissipation
`and space concerns. Because of these issues, laptop CPU’s are still typically several months behind desktop
`CPU’s in terms of processing power.
`The display is another major power consumer and again poses problems not found in a desktop machine.
`Unlike the Cathode Ray Tube (CRT) monitors used in all desktops, there are two major types of displays
`
`*This work was supported by the Hewlett—Packard Research Grants Program, the AT&T Foundation, NSF #CDA—92—14924,
`and DARPA #MDA972—95—1—0013.
`
`Page 4 of 14
`
`Page 4 of 14
`
`
`
`used in laptops — passive dual scan (STN — Super Twisted Nematic) and active matrix (TFT — Thin Film
`Transistor). The dual scan display is cheaper and easier to manufacture but has poorer picture quality,
`especially when displaying fast-moving images. The active matrix display produces excellent picture quality
`but at a higher cost and greater power consumption. Active matrix displays are also more difficult to
`manufacture and very often have several defective pixels in them. Table 1 shows some of the differences
`between typical desktop and laptop displaysl.
`
`
`
`Display
`Display size
`Weight
`Power Consumed Resolution
`Number
`
`type
`(diagonal inches)
`(lbs)
`(Watts)
`(pixels)
`of colors
`
`
`
`
`
`1280x1024
`800x600
`
`unlimited
`262,144
`
`
`
`
`
`
`
`
`
`
`
`Desktop
`Laptop
`
`1777
`11.377
`
`
`
`
`
`47.4
`1.1
`
`190 (max)
`2.7
`
`
`
`Table 1: Comparison of typical laptop and desktop displays.
`
`form factor), laptop drive design also requires increased
`In addition to reducing the physical size (i.e.
`tolerance for mechanical shocks and the ability to spin up faster than desktop drives. The latter is necessary
`because laptop drives get spun down more often in order to reduce power consumption (this is explained in
`more detail in Section 4.1). Table 2 shows the contrast between typical2 desktop and laptop drives. The
`differences in power consumption are very significant, as we will see later in this paper.
`
`
`
`Capacity
`Size
`Weight
`Power (R/ W)
`Power (Idle)
`Shock Tolerance
`
`(MBytes)
`(inchesB)
`(lbs)
`(Watts)
`(Watts)
`(Gs)
`2113
`15.0
`1.0
`7.0
`3.2
`2
`810
`8.3
`0.4
`2.1
`1.0
`100
`
`
`
`
`
`
`
`
`
`
`
`
`
` Type
`
`Desktop
`Laptop
`
`Table 2: Comparison of laptop and desktop hard drives
`
`All of the subsystems of the laptop share a single battery as their primary source of power when not
`plugged into a wall outlet. There is usually an additional small battery for the real time clock and for
`memory backup, but this is not relevant to our discussion.
`A mobile computer (for the purposes of this paper, we define a mobile computer as a laptop computer
`with wireless networking capabilities) has severe limits on its electrical power usage, and a frequent complaint
`about mobile computers is the short lifespan of the battery [11]. Battery life is rarely more than 2—3 hours for a
`heavily-used laptop. Additional features, such as larger color displays, larger and faster hard disks, powerful
`processors, more memory and CD-ROM drives are becoming common, and result in increased electrical
`power demands. Unfortunately, laptop batteries are not advancing as rapidly as the other subsystems (for a
`comparison, see Figure 1). Each new feature, unless managed properly, will only further reduce battery life
`and inhibit untethered operation.
`
`1.2 Overview
`
`In the next section we discuss laptop batteries and show why batteries are unlikely to improve significantly in
`the forseeable future. Section 3 examines relative power consumption of the major subsystems of a laptop.
`In Section 4 we survey currently applied power management techniques for each of the subsystems, and
`discuss some of the problems associated with them. Section 5 outlines several new power management ideas.
`
`1The monitor is a Nanao FlexScan T2—17TS and the laptop display is Fujitsu’s FLC29SVC6S Active Matrix LCD
`2The desktop drive is a Seagate Medalist Pro 2.1 and the laptop drive is a Seagate Marathon 810
`
`Page 5 of 14
`
`Page 5 of 14
`
`
`
`Processor (MIPS)
`Hard Disk (Capacity)
`Memory (Capacity)
`
`0
`
`1
`
`2
`
`3
`Time (Years)
`
`4
`
`5
`
`6
`
`Battery (Energy Stored)
`
`16x
`
`14x
`
`12x
`
`10x
`
`8x
`
`6x
`
`4x
`
`2x
`
`1x
`
`Improvement (compared to year 0)
`
`Page 6 of 14
`
`
`
`1995-96
`Pentium/90
`16 MB RAM
`770 MB HD
`Active Color
`
`1991-92
`386/25 MHz
`8 MB RAM
`85 MB HD
`Backlit Disp.
`
`1993-94
`486/25
`8 MB RAM
`105 MB HD
`Passive Color
`
`DISPLAY
`
`CPU
`
`HARD DRIVE
`
`70
`
`60
`
`50
`
`40
`
`30
`
`20
`
`10
`
`Percentage of Total System Power
`
`1991
`
`1992
`
`1993
`
`1994
`
`1995
`
`Year
`
`Page 7 of 14
`
`
`
`Display
`36%
`
`Wireless
`LAN
`18%
`
`Hard Drive
`18%
`
`Other
`7%
`
`CPU/Memory
`21%
`
`Page 8 of 14
`
`
`
`Powerup - Electronics activated, platters off
`Spinup - Platters start spinning
`
`Idle - Platters spinning
`
`1 2
`
`3
`
`Seek - Actual data transfer
`
`4
`5 Spindown - Platters spinning down
`6 Powerdown - Drive off
`
`2
`
`4
`
`3
`
`3.5
`
`3.0
`
`2.5
`
`2.0
`
`1.5
`
`1.0
`
`0.5
`
`Power Consumed (Watts)
`
`1
`
`0
`
`0
`
`1
`
`2
`
`3
`Time (Seconds)
`
`4
`
`5
`
`6
`
`5
`
`6
`
`Page 9 of 14
`
`
`
`Figure 4 is derived from Li, et. al’s [16] measurements and illustrates the dynamic power consumption of
`a typical laptop—optimized hard drive (a Maxtor MXL—105 III). The total energy consumed is equal to the
`entire shaded area under the curve (i.e. Energy : Watts*Seconds). The largest power drain occurs during
`spin up, shown as area 2 in the figure. Spinning up a disk requires overcoming the mechanical inertia of the
`stationary platters of a disk. Once the platters are spinning, the power required to keep them spinning is
`much lower, as shown in area 3 of the figure. Disks optimized for laptops have a shorter spin up time than
`disks intended for ordinary PC’s, to allow for frequent spin—downs to conserve energy. Of course, spinning—
`down and spinning—up a disk too frequently can result in higher overall power consumption since the energy
`required to spin up a disk is much higher than that needed to keep the disk spinning.
`In theory, the best
`power conservation happens when a disk is spun down if the energy it would spend being idle (i.e. area 3)
`is equivalent to or greater than the additional cost of spinning it back up (the area in 2). As we will see,
`this isn’t always feasible in practice.
`Research has been done on reducing the overall amount of energy used by a hard drive. This has ranged
`from simple algorithms that spin down the drive when it is idle for more than a set length of time (currently
`the most common method), to adaptive spin down techniques where the drive examines past access patterns
`to determine a dynamic spin down strategy.
`The fixed length spin down policy has one big advantage: it is very simple to implement. If the spinning
`disk is not accessed for idle_time minutes, the assumption is that there will be no disk accesses in the near
`future and the disk is spun down.
`It spins up again when there is a read/ write request. This is the only
`widely available disk management method at present. Since the user fixes the value of idlejime and rarely
`readjusts it, the savings are very limited. Setting idlejime too low results in the user waiting for the drive
`to spin up too often. Too high a value of idle_tz'me results in minimal power savings since the disk will
`remain spinning most of the time. A study by [16] has shown that the optimal value (strictly from the
`power conservation point of view) for idle_time is approximately 6 seconds. This may be ideal from the
`power perspective, but is very inconvenient for the user who will frequently have to wait for the drive to
`spin up (a spin up takes 2-6 seconds). In addition, since a hard disk is a mechanical device, it typically has
`a spin—up/spin—down life expectancy of 40,000—60,000 cycles and overly aggressive spin—down techniques will
`result in premature drive failure. For example, if idle_time is set to 6 seconds, the drive could spin—up over
`1000 times on a 5 hour cross—country flight, reducing disk life by about 2% in just one flight.
`Adaptive disk spin-down attempts to adjust to the user’s access patterns. IBM’s Adaptive Battery Life
`Extender (ABLE) [12] looks for temporal locality of reference in drive accesses to put a hard drive in a special
`idle mode that shuts down most of the electronics of the drive but does not spin-down the platters when
`accesses are not expected. The drive analyzes the frequency distribution of commands over the previous
`10—15 seconds and calculates the probability that the current command is the final one in the burst. This
`method conserves about 15% more power than a regular idle mode disk and is transparent to host software.
`Some adaptive spin-down schemes [8] propose actually spinning down the drive completely to maximize
`energy conservation, but are difficult to implement and have only been simulated so far. The caveat for each
`of these schemes is that savings can vary widely with usage. A more detailed analysis of these techniques
`can be found in [20].
`Another technique is increasing the size of the disk cache to reduce the need for spin—ups. Caching can
`improve performance, while reducing power consumption. Simulations by Douglis et. al.
`[8] show that using
`a 32 Kbyte SRAM write-buffer improves average write response by a factor of 20 or more and reduces energy
`consumption by between 15%-20%. Their simulations show that increasing the buffer beyond 32 Kbytes
`does not improve write response time, nor does it save additional energy, although this will probably vary
`with the operating system environment. Thus there is an upper bound on how useful a disk cache can be,
`and there are also negative consequences for reliability since SRAM is volatile and there is potential for data
`loss in the event of a system error.
`In summary, hard disk management is still not mature. Currently available algorithms (almost exclusively
`fixed—length spin down) can help, but are far from optimal. Adaptive algorithms are still in the preliminary
`research stage, and are difficult to implement. There are other problems inherent to hard disks 7 since they
`are mechanical devices, they have limited spin—up/spin—down cycles before drive failure.
`
`Page 10 of 14
`
`Page 10 of 14
`
`
`
`NOTE: Scale for Y-axis is logarithmic
`
`SRAM
`
`Flash Memory
`DRAM
`
`Hard Disk
`
`1995
`
`1996
`
`Year
`
`$200
`
`$100
`
`$50
`
`$20
`
`$10
`
`$5
`
`$2
`
`$1
`
`$0.5
`
`$0.2
`
`Price ($/MByte)
`
`Page 11 of 14
`
`
`
`
`
`CPU Frequency Voltage Typical Power
`Performance
`
`(MHZ)
`(Volts)
`(Watts)
`(iCOMP/Watt)
`
`Desktop 66
`5.0
`~10
`57
`
`75
`3.3
`3.0 - 4.0
`174
`
`90
`2.9
`2.5 — 3.5
`245
`
`100
`2.9
`2.0 - 3.0
`326
`
`120
`2.9
`2.5 — 3.5
`333
`133
`2.9
`3.0 - 4.0
`317
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`Table 4: Power requirements of Pentium processors for laptops.
`
`amount of power used by a circuit is proportional to the square of the voltage used, so even a small decrease
`in processor voltage results in a large decrease in the power consumed.
`The newer Pentium CPUs also have circuitry that allow the microprocessor to slow down, suspend, or
`completely shut down various subunits of the processor when they are not in use. This is transparent to
`the operating system and application software and explains the dramatic drop in power consumption for
`the more recent processors, as shown in Table 4. As power management schemes internal to the CPU start
`reaching their limits, the total power consumed will start rising, as is apparent with the fastest processor in
`the table.
`
`In addition, there are user selectable options to run the CPU at a slower speed to conserve powerithis
`is the most common user choice in most power managed laptops. The problem with user—selectable “slow”
`or “fast” CPU modes is that the user may actually end up using more power with the “slow” power—saving
`mode than by not using the power save mode at all. For example, if the user is editing a spreadsheet, having
`the CPU in its slow state is optimal, but if the user is running a calculation in the spreadsheet, having the
`CPU running slower will result in more power being used since the display and hard disk will be left on
`longer. It is impractical to expect the user to set the CPU speed manually each time so this inefficiency is
`common.
`
`4.4 System Memory
`
`One method to reduce the number of times a hard disk has to be spun up is to have a large amount
`of system memory (i.e. DRAM). This makes intuitive senseiplace the current working set in memory and
`there will be few page faults to cause the disk to spin up. Unfortunately this is not feasible in practice.
`A study by Li [17] has shown that having as little as 8 MBytes of additional DRAM can use up as much
`power as a constantly spinning hard disk. To confirm this rather surprising result, we did some calculations
`based on manufacturers data [5], and found that 8 MBytes of 60ns Extended Data Output (EDO) DRAM
`uses 2.8 W when active, compared to a typical 500 MByte hard disk which uses about 3 W. Newer memory
`technologies will probably reduce the power consumption of DRAMs, but it will still remain significant in
`comparison to a hard disk.
`This indicates that adding system memory solely to reduce disk accesses is not a workable solution. In
`fact, a user wanting to maximize battery life may need to keep system memory to an absolute minimum.
`We discuss this further in Section 5.
`
`4.5 The Display and Network Interface
`
`The display of a laptop can absorb almost half of the total available system power. Active matrix screens
`use more power than the older dual scan displays. Displays are improving rapidly in size and resolution,
`but not in terms of power consumption. Power management of displays is typically restricted to blanking
`the display after a period of inactivity. Some newer system management software allows a user to set a low
`power mode that dims the screen. Blanking the screen after a few minutes is effective in saving power but
`is not optimal.
`
`Page 12 of14
`
`Page 12 of 14
`
`
`
`Wireless network interface cards are becoming more common. The wireless Ethernet card (CSMA/ CA)
`we are using is the AT&T PC Card WaveLAN, which has a claimed consumption of 3 W during transmission,
`1.5 W when receiving and 0.2 W in sleep mode [22]. Experiments conducted on PDAs by Gauthier et. al.
`[10] support these numbers. In addition, they also noted that the time the WaveLAN takes to switch from
`sleep mode to active mode is about 100 ms 7 sufiiciently short that the user would not notice a lag if the
`card were put in sleep mode frequently.
`The wireless LAN standard (IEEE 802.11) is still being defined and will include some form of built—in
`power management when finalized. While there has been work on reducing power consumption of wireless
`network cards [10, 13], most of it is focused on a particular subsystem, not the entire mobile computer.
`
`5 Power-Conscious Memory Management
`
`Since software continues to require more memory it is useful to control the amount of memory actually
`powered up. For example, if a laptop with 40 MBytes of memory were to use only 16 Mbytes and depower
`the other 24 Mbytes, there would be very significant power savings, possibly with performance degradation.
`If a user could set (either at power—up, or dynamically) the amount of memory to be powered down, it would
`offer a method to tradeoff performance with laptop battery endurance. Since about 8 Mbytes of DRAM can
`use as much power as a spinning hard disk [17], the additional page faults (and subsequent drive spin—ups)
`would be offset by the savings from having reduced DRAM. Intel has released a new Pentium PCI chipset
`(the 82430MX PCI chipset) that has suspend and standby modes which not only put the CPU in low power
`mode, but also restrict power to system memory. The challenge is to intelligently trade power savings from
`reducing system memory against performance penalties.
`Udani [21] is researching intelligent power management where a central “Power Broker” is aware of the
`global system state and selectively shuts down laptop components based on a rule base for each group of
`applications. Applications are unmodified. Dependencies between components (e.g.
`if the the display is
`off, the hard drive can be immediately spun down) can be used to minimize power consumption without
`affecting performance.
`An idea proposed by the Video Electronics Standards Association (VESA) group is the Unified Memory
`Architecture (UMA) [4]. They propose a scheme where segments of main memory are dynamically allocated
`for video and graphics, thus eliminating the need for a separate frame buffer. This proposal is presented
`primarily as a cost saving measure, but can also be viewed from the power management point of view.
`Instead of having dedicated memory reserved for graphics (2 MB requires about 0.7 W of power), segments
`of main memory can be used as needed. This would be more efficient and flexible. For example, a word
`processing application might need only 512 KBytes whereas photo rendering may need over 2 MBytes. Each
`of these could be accommodated using the UMA scheme and the memory returned for system use after the
`application is finished. The claim by VESA is that UMA is transparent to the operating system and is
`controlled by the core BIOS logic. There are disadvantages however. Since we are using the system bus and
`system memory for all the traffic, performance degradation is likely and estimated to be between 5—15%. For
`a desktop machine this may not be acceptable, but if it can extend a laptop’s battery life by 10% there is
`strong incentive to use the scheme.
`
`6 Summary
`
`We have analyzed the various subsystems of a mobile computer from the power management perspective.
`In summary, we:
`
`1. Looked at technology trends in mobile computing
`
`2. Identified batteries as the key laggard
`
`3. Surveyed possible solutions for power management
`
`10
`
`Page 13 of 14
`
`Page 13 of 14
`
`
`
`References
`
`HAO:
`
`OONQOT
`
`
`
`
`
`Apple Computer Press Release, URL: http://www.atcon.com/maccentral/oct/bits.html
`
`Managing Mass Storage, Byte Magazine, March 1994, pp. 79783.
`
`Personal Communication, Glenn Bernasek, Email: dd314@clevelandFreenetEdu
`
`Unified Memory Architecture May Change Future PC Designs, Computer Design,
`pp. 47748, December 1995.
`
`Product Trends: High Performance DRAMs, Computer Design, pp. 1397149, December 1995.
`
`Computer Shopper, April 1995 and March 1996.
`
`Giga Information Group Survey, Computerworld, p. 1, 8 April, 1996.
`
`F. Douglis, P. Krishnan, and B. Bershad, Adaptive Disk Spin—down Policies for Mobile Computers,
`Second USENIX Symposium on Mobile and Location—Independent Computing, pp. 1217137 Ann Arbor,
`April 1995.
`
`G. Forman and J. Zahorjan, The Challenges of Mobile Computing, Technical Report UVV CSE 93-11-03,
`University of Washington, March 1994, available from fip.cs.washington.edu.
`
`P. Gauthier, D. Harada and M. Stemm, Reducing Power Consumption for the Next Generation of PDAs:
`It’s in the Network Interface!, URL: http://http.cs.berkeley.edu/ stemm/cs252/report/main/node1.html
`
`D. Freeman, Lithium-ion BatteriesiThe Next Wave, Portable Design, pp.46747, January 1996.
`
`Adaptive Power Management for Mobile Hard Drives, IBM, August 1995.
`URL: http://eagle.almaden.ibm.com/storage/oem/tech/ableback.htm
`
`T. Imielinski, S. Viswanathan and B. Badrinath Power Efiicient Filtering of Data on the Air
`URL: http://www.cs.purdue.edu:80/homes/uzk/MobileComputing.html
`
`URL: http://pentium.intel.com/procs/pentium/systems/mobile/pp90/sltech.htm
`
`URL: http://www.intel.com/IAL/powermgm/
`
`Kester Li, R. Kumpf, P Horton and T. Anderson, A Quantitative Analysis of Disk Drive Power Man—
`agement in Portable Computers, Proceedings of the 1994 Winter USENIX Conference, pp.2797291, San
`Francisco, January 1994.
`
`Kester Li, Towards A Low Power File System, Technical Report UCB/CSD 94/814, University of
`California — Berkeley, Masters Thesis, May 1994.
`
`B. Marsh and B. Zenel, Power Measurements Of Typical Notebook Computers, Technical Report MITL-
`TR-110-94, Matsushita Information Technology Laboratory, May 1994.
`
`S. Sheng, A. Chandrasekaran and R. W. Broderson, A Portable Multimedia Terminal For Personal
`Communications, IEEE Communications Magazine, pp. 64—75, December 1992.
`
`S. Udani, Power Management of Permanent Storage in Mobile Computers,
`Written Preliminary Examination Part II Report, University of Pennsylvania, April 1995.
`URL: http://www.cis.upenn.edu/~udani/papershtml
`
`S. Udani and J. Smith, The Power Broker: Intelligent Power Management for Mobile Computers,
`Technical Report MS-CIS-96-12, Department of Computer Information Science,
`University of Pennsylvania, May 1996. URL: http://www.cis.upenn.edu/~udani/papers.html
`
`
`16
`20 21
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`17
`
`18
`
`19
`
`[22] ATCT WaveLAN PCMCIA User’s Guide, October 1994.
`
`[23] Personal Communication, Clinton Winchester, US Navy. Email: winchest@oasys.dt.navy.mil
`
`11
`
`Page 14 0f14
`
`Page 14 of 14
`
`