Conference PaperPDF Available

Power and performance characteristics of USB flash drives


Abstract and Figures

Even though their capacities are still orders of magnitude lower than those of hard disks, flash storage systems are rapidly gaining importance in energy-constrained systems. This paper focuses on USB flash drives, which can provide portable storage to mobile systems or storage to systems that otherwise do not have persistent storage opportunities (e.g., low-power sensor devices). The paper presents studies relating to power consumption, energy overheads and benefits, and performance impacts of USB flash drives. The key insights obtained from these experiments are that (i) read/write costs are not significantly greater than idle costs and (ii) the size of the flash itself has only limited bearing on energy consumption.
Content may be subject to copyright.
2008 IEEE
Power and Performance Characteristics of USB Flash Drives
Kyle O’Brien, David C. Salyers, Aaron D. Striegel, Christian Poellabauer
Department of Computer Science and Engineering
University of Notre Dame
Notre Dame, IN 46556 USA
Email:{kobrie5, dsalyers, striegel, cpoellab}
Even though their capacities are still orders of mag-
nitude lower than those of hard disks, flash storage sys-
tems are rapidly gaining importance in energy-constrained
systems. This paper focuses on USB flash drives, which
can provide portable storage to mobile systems or stor-
age to systems that otherwise do not have persistent stor-
age opportunities (e.g., low-power sensor devices). The
paper presents studies relating to power consumption, en-
ergy overheads and benefits, and performance impacts of
USB flash drives. The key insi ghts obtained from t hese ex-
periments are that (i) read/write costs are not significantly
greater than idle costs and (ii) the size of the flash itself has
only limited bearing on energy consumption.
1 Introduction
Storage components of mobile wireless systems, includ-
ing laptops, handhelds, and sensor devices, are essential for
temporary and long-term preservation of information. In
the recent past, solid state, or flash storage, has emerged
as a promising alternative to hard disks. While flash mem-
ory and drives offer the potential for very low-power stor-
age, they are still significantly smaller than available disk
drives [7].
USB flash drives provide an opportunity to extend an ex-
isting mobile/wireless device with (additional) storage ca-
pacity; this is particularly important f or small low-power
devices that otherwise do not have any storage opportunities
(e.g., TelosB motes and Crossbow Stargate wireless mesh
routers). With newer areas of research such as Delay Tol-
erant Networking (DTN) [5], on-device flash capacities are
often not enough. Finally, in the absence of designing a
new device solely for their research, researchers are often
faced with employing USB flash drives to meet that need.
Unfortunately, the performance of said devices with regards
to energy and read/write speeds is poorly understood. The
usage of flash for the network researcher is especially criti-
cal as the choice of flash can have a profound effect on the
energy and performance characteristics of the overarching
system or application.
Toward this end, this paper provides a comprehensive as-
sessment of the power, energy, and performance character-
istics of flash drives. Specifically, the contributions of this
work can be summarized as follows:
Comprehensive study of USB flash drive power con-
sumption: This paper experimentally examines the
power consumption characteristics of USB flash drives
across a wide variety of considerations including man-
ufacturer, device size, operating system, read/write
blocks, and USB connection types (1.1 or 2.0).
Comprehensive performance studies: Further, this pa-
per evaluates the energy - performance tradeoff of USB
flash drives, assessing the energy costs and perfor-
mance numbers of various I/O accesses to a USB flash
2 Experimental Methodology
The measurement results were obtained using a
Fluke189 multimeter and a DSO3202A oscilloscope
(200MHz bandwidth, 2 channels, 1GSa/s sample rate). The
results were imported into and visualized using LabView.
The USB flash drives utilized in the experiments are sum-
marized in Table 1. While there are certainly commonal-
ities between different drive vendors in terms of the actual
underlying flash, our goal was to roughly assess the existing
spectrum of flash drives with selective finer grained assess-
ment within select individual vendors (Kingston, Sandisk).
The flash drives’ power characteristics were studied on
multiple host devices and operating systems:
Crossbow Stargate (Linux:) Crossbow Stargate de-
vices are frequently used as wireless mesh routers or
as powerful sensing devices (Linux 2.4.19, USB 1.1).
Manufacturer Size(s)
Axiom, Entec 1 GB
Edge, PNY 2 GB
EP 128 MB
Imation, Kanguru 512 MB
Kingston 512 MB, 1 GB, 2 GB, 4 GB
Memorex 256 MB
OCZ, Transcend 1 GB
Patriot 4 GB
Sandisk 512 MB, 2 GB, 4 GB
Simpletech, Smartmod 256 MB
Table 1. USB flash drives used in study
Mac mini (Mac OS X), Gateway Tablet (Windows XP),
Sun x2100 Workstation (Linux 2.6): For USB 2.0 per-
formance across various operating systems and sus-
pend behavior, the listed array of systems were em-
3 Power Consumption
Figure 1 captures the power consumption of the respec-
tive USB flash drives under USB 1.1 accessed via the Linux
2.4 kernel on a Stargate Crossbow device. The figure cap-
tures the average power consumed under idle conditions,
reading a significant file, and writing a significant file.
With regards to the cost of read/write over idling, the an-
swer appears to clearly be that writing to the respective flash
device when done in a bulk write (64kB blocks) does not
incur a serious cost over normal read and write operations.
The notable anomaly is the case of the Memorex flash de-
vice whereby power consumption is actually improved over
the idle case. In short, the action of reading and/or writing
for the Memorex device changes the main LED from a con-
tinual on state to one that flashes on and off, thus reducing
power significantly.
In contrast, the Sandisk devices exhibit the opposite ex-
treme of the behavior, representing a jump in total power
that cannot be solely attributed to the cost of the reading
and/or writing the device. Notably, the Sandisk devices pos-
sess the largest and brightest LEDs of the surveyed devices
representing a notable power cost for USB 1.1 interactions.
The Kingston set of devices were some of the most efficient
devices in terms of average power consumption.
While the limited impact of read and write operations
as well as the nearly indistinguishable cost of writing over
reading were certainly interesting, the most striking aspect
of the devices was the lack of engaging the low power sus-
pend mode when not in use. Per the USB specification
[1, 2], if a device does not see activity in the last 3.0 ms,
Figure 1. Power consumption of all devices -
idle, read, write under USB 1.1
the device has 7.0 ms to enter a low power suspend mode
where the device may use a maximum of 0.5 mA of cur-
rent. Devices with remote wakeup may draw up to 2.5 mA
of current. Upon further investigation of t he data line itself,
a 2-byte keep alive signal is sent to the flash drive devices
continually, thus preventing the device from ever entering
the low power mode. This keep-alive message will persist
for the duration of the host running with the only exception
being when the “Safely remove hardware” operation is per-
formed under Windows. Unmounting the drive under Linux
or ejecting the drive under Mac OS X did not push any of
the drives into suspend mode despite repeated attempts and
3.1 Examining Write Power Consumption
Figure 3 examines the effect of block size averaged
across all devices. In the figure, the same size file is writ-
ten out using different block sizes for each respective write
call. While the graph itself is not unexpected, the key transi-
tion points are interesting. A block size of 4 kB yields little
performance difference versus block sizes of up to 18 kB.
Once the 20 kB threshold is passed, the net write perfor-
mance is quickly saturated leaving minor gains (if any) for
pushing the write block size even higher. An interesting re-
search case would be to examine the effects of writing with
regards to reliability although most modern flash devices
scatter writes across the flash to avoid continually rewriting
the same area. We note that the 20 kB nicely captures the
Figure 2. Write speed - USB 1.1 - 100 MB file
Figure 3. Effect of block size USB 1.1
static content of many popular websites for the purposes of
on-device caching in the flash itself [6].
3.2 Varying Device Size
Figure 4 examines the effect of device size on the overall
power consumption of a specific series of devices, namely
variations within the Kingston devices. Due to the signif-
icantly larger power consumption of the Sandisk devices,
only the Kingston devices are highlighted in the graph. In
fact, the Sandisk devices demonstrated power savings when
switching to larger devices (i.e. gains in efficiency) over
smaller devices.
While there is a slight general increasing trend for idle
power, the read/write power consumption is relatively in-
conclusive until the 4 GB device. The cost of increasing
size appears to be largely masked and only of significant
consequence with the largest device size. Hence, from the
perspective of the network researcher, the selection of a 2
GB device has little energy cost with significant flexibil-
ity improvements over a 512 MB or 1 GB device. Further
study is warranted regarding if the trend continues for newly
emergent 8 GB and larger devices.
3.3 USB 1.1 vs. USB 2.0
Although support for USB 2.0 in small scale embed-
ded devices is less common, the increased speed for USB
2.0 (480 Mb/s) versus USB 1.1 (12 Mb/s) has significant
appeal as wireless network speeds are continually increas-
ing. While the performance of USB 2.0 is considerably
better than USB 1.1 in terms of write performance (16.97
MB/s peak write performance versus roughly 700 KB/s un-
der USB 1.1), the average power consumption is well over
double during the write operation itself as well as during
idle periods.
Table 2 illustrates the power consumption of the Patriot
4 GB drive across various combinations of platforms, oper-
ating systems, and USB connectivity. Interesting observa-
tions from the graph are that the power cost at idle time or
during a read or writ e (read is omitted due to its similar per-
formance to a write operation) are consistent despite varia-
tions in platform, CPU/motherboard, and operating system.
While this is to be expected given that USB is a well un-
derstood standard, the tightness of the numbers across plat-
forms was somewhat surprising.
Although USB 2.0 performance is significantly in-
creased over USB 1.1, the idle power consumption is quite
troubling given the fact that suspend is never engaged de-
spite long idle periods on the device itself (barring cases
where the OS hibernates or sleeps). Unless the applica-
tion strictly needs the improved performance (i.e. caching
content from an 802.11 draft n link versus an 802.11g or
802.11b link), the idle power consumption of USB 2.0
makes it quite costly relative to USB 1.1.
4 Related Work
In the direct area of USB flash drives, very little previous
work exists in characterizing the power of the devices. To
the best of our knowledge, ours is the first work to directly
assess USB flash drive performance with regards to energy
characteristics. We comment briefly on several of the more
recent related works that consider flash for the purpose of
energy savings.
The notion of energy-awareness with regards to multi-
media and mobile computing is a concept that has seen
considerable research [8]. In [7], Singleton, Nathuji, and
Schwan explore algorithms to manage flash allocation as a
Figure 4. Effect of varying flash size - USB 1.1
cache for disk access. Similarly, Bisson and Brandt in [3]
examined similar aspects to allow for improved disk spin
down for reduced energy consumption. Chen, Jiang, and
Zhang propose SmartSaver in [4] for using flash as a pre-
fetch and cache for web content for mobile nodes. These
works simply represent a small subset of the works exam-
ining the applicability of flash. Notably, the end result in
nearly of the work is that to actually provide the flash itself,
a USB flash drive is used whose characteristics are assumed
to roughly mirror actual on-device flash.
Specifically, the emergence of Delay Tolerant Network-
ing [5] as a subject of much research is especially notable.
Critically, if a researcher assumes that the flash device oper-
ate in a manner as power efficient as normal OS operations
(i.e. the drive will suspend if not used), the end result will
be a distorted picture of the true energy savings offered by
flash. Moreover, the characterization of the true costs of re-
suming and suspending the flash device make the develop-
ment of better performing suspend/resume operations a nec-
essary component for the research community as a whole.
5 Conclusion
In conclusion, this work represented one of the first
works to comprehensively address the power consumption
of USB flash drives across a multitude of combinations of
manufacturer, connecting host, and connection type. No-
tably, the most striking results are that: (i) reads and writes
while more expensive are not significantly more expensive
than normal idle operation, (ii) raw performance often does
not translate to overall energy savings (faster write perfor-
mance vs. overall energy efficiency, USB 1.1 idle vs. USB
2.0 idle, etc.).
Platform - OS USB Idle (mA) Write (mA)
Stargate /
Linux 2.4 1.1 22.37 34
x86 Workstation /
Linux 2.6 2.0 50.88 100.5
x86 Tablet /
Windows XP 2.0 50.84 105.5
PPC Mac Mini /
Mac OS X 1.1 22.4 34.5
x86 Mac Mini /
Mac OS X 2.0 50.95 110.0
Table 2. Comparison of USB 1.1 vs. USB 2.0
for various OS / platforms for the Patriot 4 GB
flash drive
This work was partially supported by the National Sci-
ence Foundation as part of an REU project through grant
CNS03-47392. Equipment support was provided in part by
a DURIP grant and a Notre Dame Faculty Grant.
[1] USB 1.1 specification. 1996.
[2] USB 2.0 specification. Apr. 2000.
[3] T. Biss on and S. Brandt. Reducing energy consumption using
a non-volatile storage cache. In Proc. of RTAS 05, 2005.
[4] F. Chen, S. Jiang, and X. Zhang. SmartSaver: turning flash
drive into a disk energy saver for mobile computers. In Proc.
of ISLPED, Tegernsee, Germany, Oct. 2006.
[5] K. Fall. A delay tolerant networking architecture for chal-
lenged Internets. In Proc. of ACM SIGCOMM, Aug. 2003.
[6] X. Li, D. Salyers, and A. Striegel. Improving packet caching
scalability through the concept of an explicit end of data
marker. In Proc. of IEEE HotWeb, Boston, MA., Nov. 2006.
[7] L. Singleton, R. Nathuji, and K. Schwan. Flash on disk for
low-power multimedia computing. In Proc. of MMCN, San
Jose, CA, Jan. 2007.
[8] F. Zheng, N. Garg, S. Sobti, C. Zhang, R. Joseph, A. Krishna-
murthy, and R. Wang. Considering the energy consumption
of mobile storage alternatives. In Proc. of MASCOTS, 2003.
... Detailed studies of low-level electrical characteristics are available [5,6,10], as well as performance studies of complete storage assemblies (e.g. SSDs) containing flash chips and controllers [1,9]. However, to the best of our knowledge, there is no experimental study to date of actual Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. ...
... The most recent of these presents uFLIP [1], a benchmark for such storage systems, with measurements of a wide range of devices; this work quantifies the degraded performance observed for random writes in many such devices. Additional work in this area includes [2] and [9]. There has been a small amount of empirical testing of raw flash chips in the wireless sensor network community [7], but this work has focused primarily on energy usage and has not addressed performance or endurance. ...
Full-text available
Reports of NAND ash device testing in the literature have for the most part been limited to examination of circuit-level parameters on raw ash chips or prototypes, and system-level parameters on entire storage subsystems. However, there has been little examination of system-level parameters of raw devices, such as mean latency and endurance values. We report the results of such tests on a variety of devices. Read, program, and erase latency were found to align closely with manufacturer's specified \typical" values in almost all cases. Program/erase endurance, however, was found to exceed specified minimum values, often by as much as a factor of 100. In addition significant performance changes were found with wear. These changes may be used to track wear, and in addition have significant implications for system performance over the lifespan of a device. Finally, random write patterns which incur performance penalties on current ash-based memory systems were found to incur no overhead on the devices themselves.
... Flashdisk was a hardware device that is widely applied by teachers to store various information or data about physical education distance learning. This is in accordance with the opinion that wireless-based storage systems are very important for the temporary and long-term preservation of information and USB flash storage has become an alternative device for storing data with low power even though it still has a small capacity (O'Brien, Salyers, Striegel, and Poellabauer, 2008). It was also supported that flash storage technology requires low energy, low prices so that it can become an economical industry, and users can replace flash drives more easily (Mordvinova, Kunkel, Baun, Ludwig, & Kunze, 2009). ...
Full-text available
The aims of study to describe the utilization of information technology and telecommunications in Physical Education (PE) distance learning process during the covid-19 pandemic. This quantitative study conducted with a survey method. The subjects of this study were primary and secondary school PE teachers who were determined based on clusters of three regions/regions, namely: schools in urban, suburban, and rural areas in the Special Region of Yogyakarta with a total of 238 people. Data was collected through a questionnaire distributed to the primary and secondary school PE teachers online and offline. The data analysis applied is descriptive quantitative with percentages. The results of the study show that PE teachers apply hardware and software that already exist and are popular in Indonesia. PE teachers communicate with their students, especially WhatsApp, Email, Instagram, and Facebook, and also use social media applications in various ways. The utilzation of the Google Classroom as a Learning Management System (LMS) is very dominant.
... EEPROMs usually offer superbly low current levels but are limited to memory capacities in the MB range. USB drives offer memory capacities up to 100's of GB, but use a considerable amount of power; typically drawing 20-40 mA when idle [10]. SD cards are the best of both worlds, with GB saving capacities, fast saving times in the mA range and are capable of begin reduced to µA power consumption when not in use. ...
Full-text available
In many areas of science, Arduino based data loggers have become common enabling instruments because of their low cost and ease of use. However, battery life is commonly the limiting factor - particularly in respect of writing data to embedded SD cards. In this paper, various methods by which to optimise an SD card based data logger using an Arduino UNO, Atmega328P at 5 V at 16 MHz and the Atmega328P at 3.3 V at 8 MHz is explored. With the bare Atmega328P chip in sleep mode, the lifetime of a 2400 mAH battery can theoretically exceed 10 years, although this is reduced to only 3 months following the introduction of an SD card. The exact power consumption of an Arduino/SD card during saving events is analysed for the first time and is found to take up to 200 ms with current spikes up to 80 mA for every initialisation and saving event dramatically increasing the average current consumption of fast data loggers. Through the use of a power control MOSFET with proper initialisation and timing of SD saving events, it is found that the Atmega328P can be set up to measure data once every two seconds whilst also ensuring a battery lifetime of one year. With the novel techniques presented here, a new method for maximising the lifetime of Atmega328P microcontroller circuits for environmental data logging applications has been achieved; allowing researchers to record data using a cheap and reproducible system.
... The most recent of these presents uFLIP [7], a benchmark for such storage systems, with measurements of a wide range of devices; this work quantifies the degraded performance observed for random writes in many such devices. Additional work in this area includes [14], [27], and [1] Ben-Aroyo and Toledo [5] have presented detailed theoretical analyses of bounds on wear-leveling performance; however for realistic flash devices (i.e. with erase block size > 1 page) their results show the existence of a bound but not its value. ...
Conference Paper
Full-text available
We examine the write endurance of USB flash drives using a range of approaches: chip-level measurements, reverse engineering, timing analysis, whole-device endurance testing, and simulation. The focus of our investigation is not only measured endurance, but underlying factors at the level of chips and algorithms--both typical and ideal--which determine the endurance of a device. Our chip-level measurements show endurance far in excess of nominal values quoted by manufacturers, by a factor of as much as 100. We reverse engineer specifics of the Flash Translation Layers (FTLs) used by several devices, and find a close correlation between measured whole-device endurance and predictions from reverse-engineered FTL parameters and measured chip endurance values. We present methods based on analysis of operation latency which provide a non-intrusive mechanism for determining FTL parameters. Finally we present Monte Carlo simulation results giving numerical bounds on endurance achievable by any on-line algorithm in the face of arbitrary or malicious access patterns.
... Simply insert the pen drive inside the USB ports in-built in the system, for security against virus just scan the pen drive and view or work pen drive from the my computer options or through short cuts. The advantages of using such USB pen drives and flash drives are mainly because their power consumption and energy overhead features [3]. The one more reason for the popularity is the mobility of the device and the ideal cost is also a main feature. ...
This paper proposes an extremely innovative and creative pen-drive model which is totally computerized. The Imagination of a pen drive with a display and in-build USB ports has resulted in the development of this paper. The overall idea of this proposal is to build a pen drive which can be used even without a computer system. The challenge in this process is to develop an operating system to the pen drive device which is capable of displaying the contents of the pen- drive and it must have the capacity to share its data through transmissions to other pen drive by means of the USB ports in-build in it. The important thing to be taken in consideration is the cost efficiency of the pen drive device and the prevention of the attack of the virus on the data a device to make it more reliable and secure resource through which we can do all the operation like pen drive but even without the use of the system. and the inbuilt USB ports which helps in converting a pen drive into a Computerized Pen Drive with amazing features in it. The rate of the pen drives which is implemented with these features can also be cheap as the normal drives being in the market now. The cost mainly depends on the memory capacity and the quality of the display required. This can really create a dynamic storage and transmission device which can be used for both storage and also transmission of data directly even without a computer system which sounds brilliantly and will provide a way to innovate a new gadget. The important blocks to be implemented in order to enable the design of the Computerized Pen Drive, we need to focus primarily on the creation of an operating system to make the display work and the transmission of the data from other devices or to the other pen drives from this Computerized Pen Drive is possible only with the support of the an operating system. Designing the operation system alone is not enough to process this Computerized Pen Drive, we need a processor which is to be built up inside the pen drive only then the requests from the users can be processed and this Computerized Pen Drive can work perform the operation efficiently. The storage medium is also necessary for the device and the capacity of the storage depends on the cost. Designing the storage medium can be done using the same concepts used in a normal pen drive. Concentration on the measures which are to be taken to prevent the Virus attack is also an important thing to be considered while designing. Because the virus is a major problem when working with the pen drive, it can just corrupt the entire data present in the pen drive. Implementing a pen drive with all the features and with all the functional units as mentioned so far is surely a tedious process but the success lies overcoming all the challenges and creating a innovative gadget Computerized Pen Drive. This device will be handy and compact with an ease to operate anywhere even without the computer system.
... A number of recent benchmarking studies have examined performance characteristics of flash devices [25]- [27]. One extensive study is uFLIP [7], a benchmark for understanding flash device performance under various IO patterns; measurements of a wide range of devices are presented. ...
Conference Paper
Full-text available
There is a wide gap between the potential performance of NAND flash-based solid state drives (SSDs) and their performance in many real-world applications; understanding this gap requires knowledge of their behavior and internal algorithms for various workloads. We develop analytic models for two commonly-used Flash Translation Layer (FTL) algorithms, as used in SSDs, as well as a methodology for applying these models to real-world workloads. We demonstrate the accuracy of these models via simulation, extend this approach to incorporate measurement-based approximations when detailed parameters are unknown, and validate this methodology against real devices.
A compact, low-cost, and highly efficient wireless power transfer (WPT) system is presented. Besides transferring power, this system provides the required bandwidth for data communication; thus, it can be exploited as a wireless digital connector. It utilizes two off-the-shelf cylindrical dielectric resonators with a dielectric constant of 33 and a loss tangent of 2.5×10-4. The mutually coupled dielectric resonators operate in their magnetic-dipole mode. The resonance frequency of this mode determines the optimum operating frequency of the system which is approximately 7 GHz. The power transfer efficiency of more than 43% is maintained for distances up to four times the resonator radius. The maximum power transfer efficiency as a function of distance and misalignment is computed and experimentally verified. Its relative bandwidth renders the proposed WPT system appropriate for high-rate data transmission.
Conference Paper
Message sequence charts (MSCs) and their higher-order formalism in terms of message sequence graphs (MSGs) provide an intuitive way to describe communication scenarios. Naturally, quantitative aspects such as the probability of failure, maximal latency or the expected energy consumption play a crucial role for communication systems. In this paper, we introduce quantitativeMSGs with costs or rewards and stochastic timing information in terms of rates. To perform a quantitative analysis, we propose a branching-time semantics for MSGs as possibly infinite continuous-time Markov chains (CTMCs) interpreting delayed choice on the partial-order semantics of MSGs. Whereas for locally synchronized MSGs a finite-state bisimulation quotient can be found and thus, standard algorithms for CTMCs can be applied, this is not the case in general. However, using a truncation-based approach we show how approximative solutions can be obtained. Our implementation shows feasibility of this approach exploiting reliability, resilience and energy consumption.
Conference Paper
SSD (Solid State Disk) is reconsidered as the next storage device, an alternative to the HDD (Hard Disk Driver). The read/write performance and energy consumption are main aspects to the users. In our experiment, we recognize that the performance and energy consumption of SSD, based on NAND Flash, are mostly related with the file system and block I/O scheduler. In order to gain higher performance and lower energy consumption, we test the different combination of file system and scheduler under workload simulator, File bench, using three kinds of commercial SSDs. According to the different combination of file system and block I/O, we analyze the performance parameter, IOPS, and energy consumption parameter, POWER, under some special workload. Lastly, we present a parameter, aggregative indicator (AI), to evaluate the overall characteristic of some combination of file system and block I/O, which synthesizes IOPS and POWER. It is to find a better combination for special workload. In the experiment, the combination of extent file system (ext2 or ext3) and CFQ expresses better more value of the aggregative indicator than others.
Full-text available
Over the past few years, the Web has witnessed an explosion of dynamic content generation to provide Web users with an interactive and personalized experience. While traditional Web caching techniques work well when redundancy occurs on an object-level basis (page, image, etc.), the use of dynamic content presents unique challenges. Although past work has addressed mechanisms for detecting redundancy despite dynamic content, the scalability of such techniques is limited. In this paper, we present a technique for explicit packet boundary delineation to enable scalable and highly efficient packet caching in the network. Our approach, explicit end of data (EEOD), does not require client-side modification and requires only minimal server-side modifications. We demonstrate through experimental studies on an Apache Web server 25% and 30% relative improvements in terms of bandwidth efficiency and retrieval time over current approaches in the literature
Full-text available
We can outperform the most powerful disk spin-down algorithm available by using a small non-volatile storage write-only cache to temporarily store hard disk write re-quests while a disk is spun-down. In our experiments, we use a USB thumb drive device as our non-volatile storage cache. By redirecting writes to a flash memory while a disk is spun-down we avoid costly hard disk cycle start-stop op-erations, thus increasing hard disk reliability and reducing energy consumption. We also show that such a storage cache is necessary to provide feasible performance from a disk spin-down al-gorithm in a general purpose operating environment. The problem arises from typical general purpose operating sys-tem applications generating periodic disk activity while a system is not in use. By using small, cheap, and non-volatile flash memory we can improve the energy savings of a highly adaptive machine learning disk spin-down algo-rithm by 36%, and the 15 and 45 second fixed-timeout al-gorithm by 57% and 46%, respectively. We can also reduce the relative fraction of hard disk "contact start-stop cycles" by 80% for the DSA algorithm, and with the 15 and 45 sec-ond fixed-timeout algorithm 79% and 73%, respectively.
Mobile multimedia computers require large amounts of data storage, yet must consume low power in order to prolong battery life. Solid-state storage offers low power consumption, but its capacity is an order of magnitude smaller than the hard disks needed for high-resolution photos and digital video. In order to create a device with the space of a hard drive, yet the low power consumption of solid-state storage, hardware manufacturers have proposed using flash memory as a write buffer on mobile systems. This paper evaluates the power savings of such an approach and also considers other possible flash allocation algorithms, using both hardware-and software-level flash management. Its contributions also include a set of typical multimedia-rich workloads for mobile systems and power models based upon current disk and flash technology. Based on these workloads, we demonstrate an average power savings of 267 mW (53% of disk power) using hardware-only approaches. Next, we propose another algorithm, termed Energy-efficient Virtual Storage using Application-Level Framing (EVS-ALF), which uses both hardware and software for power management. By collecting information from the applications and using this metadata to perform intelligent flash allocation and prefetching, EVS-ALF achieves an average power savings of 307 mW (61%), another 8% improvement over hardware-only techniques.
Conference Paper
The highly successful architecture and protocols of today's Internet may operate poorly in environments characterized by very long delay paths and frequent network partitions. These problems are exacerbated by end nodes with limited power or memory resources. Often deployed in mobile and extreme environments lacking continuous connectivity, many such networks have their own specialized protocols, and do not utilize IP. To achieve interoperability between them, we propose a network architecture and application interface structured around optionally-reliable asynchronous message forwarding, with limited expectations of end-to-end connectivity and node resources. The architecture operates as an overlay above the transport layers of the networks it interconnects, and provides key services such as in-network data storage and retransmission, interoperable naming, authenticated forwarding and a coarse-grained class of service.
Conference Paper
In a mobile computer the hard disk consumes a considerable amount of energy. Existing dynamic power management policies usually take conservative approaches to save disk energy, and disk energy consumption remains a serious issue. Meanwhile, the flash drive is becoming a must-have portable storage device for almost every laptop user on travel. In this paper, we propose to make another highly desired use of the flash drive saving disk energy. This is achieved by using the flash drive as a standby buffer for caching and prefetching disk data. Our design significantly extends disk idle times with careful and deliberate consideration of the particular characteristics of the flash drive. Trace-driven simulations show that up to 41% of disk energy can be saved with a relatively small amount of data written to the flash drive.
Conference Paper
This paper is motivated by a simple question: what are the energy consumption characteristics of mobile storage alternatives? To answer this question, we are faced with a design space of multiple dimensions. Two important dimensions are the type of storage technologies and the type of file systems. In this paper, we explore some options along each of these two dimensions. We have constructed a logical-disk system, which can be configured to run on different storage technologies and to emulate the behavior of different file systems. As we explore these configuration options, we find that the energy behavior is determined by a complex interaction of three factors: the power management mechanism of the storage device, the distribution of idleness in the workload, and the file system strategies that attempt to exploit and bridge these first two factors.
Increasingly, network applications must communicate with counterparts across disparate networking environments characterized by significantly different sets of physical and operational constraints; wide variations in transmission latency are particularly troublesome. The proposed Interplanetary Internet, which must encompass both terrestrial and interplanetary links, is an extreme case. An architecture based on a "least common denominator" protocol that can operate successfully and (where required) reliably in multiple disparate environments would simplify the development and deployment of such applications. The Internet protocols are ill suited for this purpose. We identify three fundamental principles that would underlie a delay-tolerant networking (DTN) architecture and describe the main structural elements of that architecture, centered on a new end-to-end overlay network protocol called Bundling. We also examine Internet infrastructure adaptations that might yield comparable performance but conclude that the simplicity of the DTN architecture promises easier deployment and extension.