DOC PREVIEW
MIT 6 893 - Power and Performance Analysis of PDA Architectures

This preview shows page 1-2-3-4 out of 11 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Power and Performance Analysis of PDA ArchitecturesRobert LeeRipal NathujiDecember 12, 2000AbstractAs Personal Digital Assistant (PDA) usage increases, so does the demand for better performance whileadhering to strict power consumption limitations. As of the time of this study, the majority of PDAusage involves user-level applications which are not yet as computationally intensive as their desktop PCworkstation counterparts. Because of that, it seems inappropriate to apply common benchmarks to thesearchitectures, as it would not yield a particularly relevant analysis of how well the performance of thesemachines meets with current needs. Accordingly, new benchmarks specifically targetting current usagepatterns must be devised. By analyzing the performance patterns and power usage footprint of codes thatspecifically target common user tasks and hardware subsystems (such as LCD display, audio, etc) one canisolate areas of weakness in present architectures.IntroductionThe growing importance and prevelance of mobile computing devices underscores the need for low-powercomputing. Increased dependence on the battery forces hardware designers to both use less power and beable to satisfy increasingly intensive performance requirements. To this end, PDA architechtures cannotbe accurately judged by existing standards. The aim of this paper is to present and employ a standardmethodology for evaluating PDA’s that is more fitting to the specific constraints that they face.Methodology1. ApproachBased on usage patterns we chose to isolate three areas to examine for performance. Because of the currentniche that the portable devices that we are examining hold, it is imperative that they be capable of relativelyfine-grained graphics. Consumers expect a high-level graphical user interface (GUI) similar to MicrosoftWindows. This is evidenced by the fact that most PDAs are currently sold running the Microsoft WindowsCE operating system. Because of that, we decided that the CE operating system was a good platform onwhich to conduct our benchmark measurements because it both reflected current usage, and would allow fora standard platform to be used across our samples.The four PDAs that we chose to investigate are the Compaq Aero, the Compaq iPAQ, the HewlettPackard Jornada, and the Casio Cassiopeia. The Aero is powered by the MIPS R4000 processor at a rate of70 MHz and 16 MB of RAM. The iPAQ is powered by an Intel StrongARM processor at a rate of 206 MHzand 32 MB of RAM. The Cassiopeia is powered by a MIPS VR4121 processor at a rate of 131 MHz and 32MB of RAM. Finally, the Jornada is an SH3 based PDA clocked at 133 MHz and has 32 MB of RAM.To reflect the importance of supporting such a GUI, we decided to write a benchmark to measure theperformance(speed) of running common windowing tasks. This includes repositioning windows, toolbars andother elements as well as repainting and refreshing the screen.Another important usage pattern that we noted was the high rate of task switching. Because of theconsumer applications that are being used on these machines, it is often necessary to switch from oneapplication, to another. For example, in the course of one logical transaction, it may be necessary to enterthe “tasks list” and then switch to the “calendar” and then perhaps to the “address book” all in the courseof jotting down a quick business meeting. To this end, we wrote another benchmark program that selectsfrom the list of active threads running in CE and cycles between them an arbitrarily large number of times.The aim of this is to measure the speed of the context switch.Lastly, with the growing importance of supporting Java applications especially in embedded systems,we thought it would be appropriate to rate the performance of exisiting JVM’s on our PDA samples. Wechose a popular VM (and more importantly one that supported all of our architectures) called CrEme fromNSIcom. By measuring the performance of the VM by using an exisiting benchmark suite, selected codesfrom SPECjvm98 trimmed down to meet memory limitations, we aim to evaluate how well each PDA isequipped to run the increasing number of java codes.The other aspect of PDA’s under investigation was power. This includes both high level power mea-surements during power on and other runtime modes, as well as lower level measurements to estimate thepower consumption of specific machine code sequences. These measurements were taken directly from theDC power source to the PDA’s. Our choice of methodology carried both advantages and disadvantages. Ourinitial goal was to be able to observe the changes in power consumption at the instruction level granularity.Unfortunately, due to the intrinsic nature of electronic devices to act as lowpass filters, we could not obtainprecise changes in power consumption during runtime. On the otherhand, our approach provides an acuratepicture of the power consumption of the PDA package as a whole. This is appropriate because although1the details of the processor architecture are important to the performance of the PDA, this study takes intoconsideration the ability of commercial PDAs in their enirety.Results1. PerformanceThe performance benchmarks that we developed for use across the PDAs differ from the standard set ofscientific application benchmarks. The motivation for deviating from the standard toolkit of benchmarksis an analysis of usage patterns [HN98]. By developing this set of benchmark codes with portablility andcross-platform compatibility in mind as well as some notions of consumer usage, we aimed to accuratelycritique the ability of current commercially available hardware to satiate current consumer usage patterns.To that end, we introduce four performance metrics by which to consider these handheld platforms, twocustom applications aimed to heavily exercise the ability of the platform to perform task switching and GUIwindowing operations.Given what we see as a trend in the computer industry today, we feel that increasingly, Sun’s Javatechnology is playing an increasingly important role in computing, and specifically in embedded systems. Inexamining the plausibility of supporting Java codes on PDAs, we decided to use SPECjvm98 as a metric forcomparison. Using a popular Virtual Machine (VM) implementation called CrEmE, we ran selected codesfrom the SPEC suite and compared the performance on these codes across the different machines. The codesthat we selected were jess, jack and mpeg.


View Full Document

MIT 6 893 - Power and Performance Analysis of PDA Architectures

Documents in this Course
Toolkits

Toolkits

16 pages

Cricket

Cricket

29 pages

Quiz 1

Quiz 1

8 pages

Security

Security

28 pages

Load more
Download Power and Performance Analysis of PDA Architectures
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Power and Performance Analysis of PDA Architectures and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Power and Performance Analysis of PDA Architectures 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?