SYNCROD: AN ENERGY-MINIMIZATION METHOD

FOR DISTRIBUTED FILE SYNCHRONIZATION ALGORITHMS

ON MOBILE DEVICES

Zach Rattner

, Hamilton Turner

, Danny Guymon

, Brian Dougherty

, Jules White

and Sean Eade

Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, Virginia, U.S.A.

Siemens Corporate Research, Princeton, New Jersey, U.S.A.

Keywords:

Energy Optimization, File Synchronization, Mobile, Wireless Networking.

Abstract:

As mobile computing platforms become ubiquitous, the need to keep data synchronized between multiple

devices becomes increasingly common. However, mobile devices have limited battery capacity, and ﬁle syn-

chronization requires extensive use of power-hungry network interfaces. This paper introduces Syncrod, an

approach for optimizing ﬁle synchronization algorithm parameters in order to minimize total energy consump-

tion. This paper presents a formal model for describing mobile ﬁle synchronization energy consumption, and

an example of ﬁtting a general ﬁle synchronization algorithm to the provided model. Empirical results of

running an energy optimization on a general ﬁle synchronization algorithm are shown to provide rapid ﬁle

transfer while using near 0.003% of a standard smartphone battery energy per MB. The provided approach

can be used to ﬁnd the the most energy-efﬁcient parameters for any ﬁle synchronization algorithm that can be

ﬁt into the provided formal model.

1 INTRODUCTION

Gartner research shows that smartphone sales have

increased 96% from 2009 to 2010(Gartner, 2010).

As smartphones continue to grow in popularity, con-

sumers are creating, editing, and sharing data from

many computing devices, such as cellular devices,

laptop computers, and tablets. This surge in the view-

ing and editing of content from both mobile devices

and home computers has increased the need for dis-

tributed ﬁle synchronization software. As ﬁles are

edited on a mobile device or home computer, dis-

tributed ﬁle synchronization software automatically

transfers the changes between devices so that all com-

puters have an up-to-date copy of the data.

Distributed ﬁle synchronization approaches trans-

fer data using the wireless network interfaces, such as

Wi-Fi and 3G on a mobile device, which consumes a

signiﬁcant amount of energy (Lagerspetz et al., 2010;

Mummert et al., 1999). Energy consumption is a criti-

cal concern on mobile devices due to their limited bat-

tery capacity (Agarral et al., 2002; Balasubramanian

et al., 2009; Mudge, 2001; Shye et al., 2010). Prior

work on distributed ﬁle synchronization has focused

on optimizing synchronization quality of service

(QoS) properties, such as end-to-end synchronization

time.

Open Problem ⇒ Determining how to Mini-

mize Power Consumption of Distributed File Syn-

chronization Algorithms for Mobile Devices.

Due to the high importance of energy consump-

tion on mobile devices, there is a need for approaches

to aid in optimizing distributed ﬁle synchronization

algorithms to consume less energy on mobile devices.

Distributed ﬁle synchronization algorithms rely on a

number of conﬁgurable parameters, such as the block

and hash size used to track and transfer changes,

which have a direct impact on power consumption.

Moreover, depending on the context of a device (e.g.,

on a 3G or Wi-Fi network), the optimal values for

these parameters vary.

Solution Approach ⇒ A Modeling and Opti-

mization Approach for Deriving Distributed File

Synchronization Parameter Values to Minimize

Energy Consumption.

In order to address the challenges of deriv-

ing algorithm parameters that result in lowest over-

all energy usage, this paper presents Syncrod i.e.

synchronization daemon. Syncrod is a formal model

for estimating the energy consumption of a distributed

135

Rattner Z., Turner H., Guymon D., Dougherty B., White J. and Eade S..

SYNCROD: AN ENERGY-MINIMIZATION METHOD FOR DISTRIBUTED FILE SYNCHRONIZATION ALGORITHMS ON MOBILE DEVICES.

DOI: 10.5220/0003822601350141

In Proceedings of the 2nd International Conference on Pervasive Embedded Computing and Communication Systems (PECCS-2012), pages 135-141

ISBN: 978-989-8565-00-6

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

ﬁle synchronization algorithm with varying block,

hash size, compression ratio, and other algorithmic

parameters. Syncrod also provides an optimiza-

tion method that can be used to automatically se-

lect ﬁle synchronization algorithm parameters, such

as ﬁle compression ratio, in order to assist distributed

ﬁle synchronization algorithm developers in creating

energy-efﬁcient ﬁle synchronization algorithms. Syn-

crod uses a formal model of data synchronization out-

lined in Section 3.

This paper provides the following contributions

to the study of distributed ﬁle synchronization algo-

rithms optimized for mobile applications:

• This paper presents a formal model, built from

empirical data, for estimating the energy con-

sumption of distributed ﬁle synchronization algo-

rithms.

• This paper describes Syncrod, a method for de-

riving the algorithm parameters values for a dis-

tributed ﬁle synchronization algorithm to mini-

mize energy consumption.

• This paper presents an example of using the

Syncrod optimization method to locate the most

energy-efﬁcient algorithm parameters for a spe-

ciﬁc distributed ﬁle synchronization algorithm.

The remainder of this paper is organized as fol-

lows: Section 2 describes the challenges that make

it difﬁcult to derive the distributed ﬁle synchroniza-

tion algorithm parameters that result in minimal total

energy usage; Section 3 formally deﬁnes the energy

used during the ﬁle synchronization process; Sec-

tion 4 empirically evaluates the impact of Syncrod on

execution time and energy consumption; Section 5

presents related work; and Section 6 presents con-

cluding remarks and lessons learned.

2 CHALLENGES OF

DETERMINING

ENERGY-EFFICIENT

PARAMETER COMBINATIONS

FOR FILE SYNCHRONIZATION

ALGORITHMS

Determining the most energy-efﬁcientcombination of

algorithm parameters for a distributed ﬁle synchro-

nization algorithm is challenging due to the difﬁculty

of understanding how algorithm software parameter

values impact power consumption. Algorithm param-

eters values have complex compounding effects upon

lower software layers, such as the operating system,

network stack, and device drivers. Moreover, dy-

namically changing mobile environment properties,

such as available signal strength or network types,

can cause signiﬁcant changes in synchronization al-

gorithm operation, thereby changing the parameter

values that will result in minimal power consumption.

The remainder of this section describes, in detail, the

challenges of identifying the ﬁle synchronization al-

gorithm parameter values that will result in minimal

energy consumption for various environments.

2.1 Challenge 1: Determining the

Impact of File Synchronization

Parameters on Energy

Consumption is Difﬁcult due to

Software Abstraction

In a distributed ﬁle synchronization algorithm for mo-

bile devices, the algorithm’s input parameters are the

primary method of impacting total energy consump-

tion. For example, increasing the amount of com-

pression applied before transmitting the ﬁle across a

network interface may increase the total CPU time

while decreasing the total transmission time. Both

CPU time and wireless data transfer impact power

consumption, and thus precisely quantifying the im-

pact of these types of trade-offs on energy consump-

tion is hard.

Similarly, there are many layers of software

abstraction between developer-written software and

energy-consuming hardware, and there are often no

built-in methodsfor tracking the energy used by a spe-

ciﬁc piece of software. Additionally, run-time events,

such as a user choosing to synchronize a large au-

dio ﬁle, can result in different software branch exe-

cutions with different energy consumption character-

istics. This hidden complexity between algorithm pa-

rameters and power consumption, coupled with the

possibility of run-time events, makes it difﬁcult to de-

termine what algorithm parameter values will result

in minimal total energy usage. Section 3.3 identiﬁes

multiple input parameters that can be used to modify

the behavior and energy consumption of a distributed

ﬁle synchronization algorithm, and Section 3.2 de-

rives general equations useful for measuring energy

consumption of each stage of a distributed ﬁle syn-

chronization algorithm.

PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems

136

2.2 Challenge 2: Discerning the Most

Energy-efﬁcient Network for a

Speciﬁc Context is Difﬁcult

Mobile devices transmit and receive data over a vari-

ety of wireless network types, such as Wi-Fi, 3G, and

Bluetooth. A key complexity for choosing parameter

values to minimize energy consumption is that each

type of network has a different energy consumption

proﬁle that is most power efﬁcient for a speciﬁc type

of data transfer.

For example, if a large ﬁle is being synchro-

nized for the ﬁrst time, an energy consuming high-

bandwidth network interface, such as Wi-Fi, may ac-

tually require less overall energy consumption than a

lower-bandwidth and less energy consuming alterna-

tive, such as 3G. Energy consumption is calculated as

the instantaneous energy consumption of the network

interface multiplied by the amount of time taken to

transfer the data. For large ﬁles, the increase in trans-

fer time of the lower-bandwidth 3G interface can end

up in a higher total energy consumption versus Wi-Fi.

Conversely, if a small amount of changed data is be-

ing transmitted, then a low-power low-speed network

may be appropriate.

3 FORMAL DEFINITION OF

ENERGY CONSUMPTION IN

DATA SYNCHRONIZATION

ALGORITHMS

To address the challenges described in Section 2,

we developed a formal method for estimating energy

consumption of distributed ﬁle synchronization algo-

rithms. This model allows our optimization method

to be easily applied to other work in the ﬁeld of

distributed ﬁle synchronization. The overall energy

equations presented here be used as the objective

functions in Section 4 to derive parameter values that

minimize energy consumption.

3.1 Syncrod & Power Optimization

Overview

Creating a method of deriving the most power-

efﬁcient parameters for a distributed ﬁle synchroniza-

tion algorithm requires an in-depth understanding of

how ﬁle synchronization components impact power

consumption on a mobile device. This understand-

ing is hard to attain, due to the vast amount of com-

plexity encapsulated between application-level soft-

ware and power-consuming hardware. This signiﬁ-

cant complexity prompted the use of a formal model

that would provide an objective function capable of

being optimized. Section 1 presents a formal model

for distributed ﬁle synchronization energy consump-

tion. The completed formal model of the distributed

ﬁle synchronization algorithm described in Section

3.2 is empirically evaluated in Section 4 to deter-

mine the algorithm parameters that resulted in the

least overall energy usage. The algorithm parameters

derived show that the algorithm presented in 3.2 is

able to transfer a 1MB ﬁle using less than 0.003% of

a standard smartphone battery capacity. The formal

model presented can be directly used with other dis-

tributed ﬁle synchronization algorithms and use the

optimization methods presented in Section 4 to derive

energy-efﬁcient values for their algorithm’s parame-

ters.

3.2 Formal Deﬁnition of Data

Synchronization Process

Figure 1: Overview of upload procedure on mobile device.

An example distributed ﬁle synchronization algo-

rithm is presented here for motivation. This algorithm

is outlined in Figure 1. The algorithm operates by ﬁrst

compressing large ﬁles, and then breaking these com-

pressed ﬁles into blocks during the partitioning stage.

In the initial synchronization, all of these blocks are

sent across the network to the remote device. In sub-

sequent synchronizations, only the blocks that have

modiﬁcations must be sent across the network. The

receiving device can ‘stitch’ the received block into

its existing set of blocks to reconstitute the entire ﬁle

and then decompress the completed set of blocks to

form the original ﬁle. We use the following subscripts

to refer to speciﬁc steps in the synchronization algo-

rithm:

• com, compression stage,

• part, partitioning stage,

• up, upload stage,

• down, download stage,

• rec, reconstitution stage,

SYNCROD: AN ENERGY-MINIMIZATION METHOD FOR DISTRIBUTED FILE SYNCHRONIZATION

ALGORITHMS ON MOBILE DEVICES

137

• dec, decompression stage.

3.2.1 Total Energy Usage

The total energy usage e can be deﬁned as the sum of

the energies used at each step of the algorithm, e.g.

the energy used during compression e

com

, the energy

used during partitioning e

part

, etc. The equation for

total energy is deﬁned as follows:

e(F, C, B, H, N) = (1)

com

(F, C) + e

part

(C, B) + e

(F, B, H, N)+

down

(F, B, H, N) + e

rec

(F, C) + e

dec

(F, C)

3.2.2 Total Energy Represented as Power over

Time

It makes sense to represent the total energy usage

in terms of power draw, by integrating power in

each stage of the synchronization algorithm over the

amount of time spent in that stage, as shown in Equa-

tion 2.

e(F, C, B, H, N) = (2)

com

(F,C)

com

(t)dt +

part

(C,B)

part

(t)dt+

(F,B,H,N)

dt +

down

(F,B,H,N)

down

dt+

rec

(F,C)

rec

(t)dt +

dec

(F,C)

dec

(t)dt

If we can derive the speciﬁc power equations and

time values for a distributed ﬁle synchronization algo-

rithm, then this high-level energy function is usable as

the objective function for minimization optimization.

3.2.3 Execution Time

The overall time taken by the synchronization algo-

rithm is the sum of the time taken by each stage of

the algorithm. For simpliﬁcation, we assume that the

uploading and downloading (t

and t

down

) are com-

pletely separate stages and do not occur simultane-

ously. More complex equations that model simultane-

ous upload/download can be deﬁned in future works.

t(F, C, B, H, N) = (3)

com

(F, C) + t

part

(C, B) + t

(F, B, H, N)+

down

(F, B, H, N) +t

rec

(F, C) + t

dec

(F, C).

3.3 Distributed File Synchronization

Design Decisions that Impact

Energy Consumption on Mobile

Devices

The purpose of our optimization approach was to ﬁnd

values for the following algorithm parameters that

resulted in minimal energy consumption due to dis-

tributed ﬁle synchronization:

• B - The size of each data block size, in bytes. Be-

cause only complete blocks are sent across the

network, overly large values for B will result in

extraneous network trafﬁc when only a few bits

of a ﬁle are changed, and the extra network trafﬁc

will result in increased energy consumption. Con-

versely, overly small values of B will result in a

need to transfer many hash values across the net-

work, again resulted in extraneous network trafﬁc.

• H - The size of each data block hash, in bytes.

Overly small values for H will increase the prob-

ability of a hash collision, which would fool the

synchronization algorithm into believing a ﬁle has

not been changed when in fact is has. Overly large

values for H result in increased network trafﬁc

and increased CPU load, both of which increase

energy consumption.

• C - Represents the amount of time and processing

power that will be used in attempting to compress

the ﬁle. Overly large values for C will increase

CPU load when compressing and decompressing

ﬁles, while values of C that are too small will re-

sult in increased network trafﬁc. Both of these

error types can increase energy consumption.

• N - The network type to use to transmit the data.

The chosen network has a signiﬁcant effect on

both data transfer speed and energy used to trans-

mit the ﬁles.

Additionally, F represents the ﬁle size, in bytes,

of the ﬁle to be synchronized. This parameter is not

a design variable, since the algorithm has no choice

over the ﬁle that will be synchronized. However, a

value for F is required to calculating the values of the

objective function e.g. Equation 2.

PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems

138

4 USING Syncrod TO LOCATE

DISTRIBUTED FILE

SYNCHRONIZATION

ALGORITHM PARAMETERS

THAT RESULT IN MINIMAL

ENERGY USAGE ON MOBILE

DEVICES

We attempted to minimize the time spent in each of

the six sections of the distributed ﬁle synchronization

algorithm. Since energy consumption is the result of

power usage over time, we reasoned that minimizing

the time taken would result in lower energy usage.

4.1 Experimental Platform

The platform used to execute these experiments was a

Dell Inspiron 530 with Intel Core 2 Duo E7200 exe-

cuting at 2.53GHz, with 1 GB of memory and 32 GB

of available disk space.

4.2 Derivation of Algorithm-speciﬁc

Optimization Functions

In order to leverage the Syncrod formal model for

estimating energy consumption, a developer must ﬁt

speciﬁc power and time equations into the objective

function in Equation 2. Due to space constraints we

cannot outline all of our synchronization algorithm-

speciﬁc functions, but we emperically measured our

algorithm’s time and power consumption for each

stage of synchronization across a range of ﬁle types

and sizes.

4.3 Experiment 1: Execution Time

Minimization with Syncrod

As ﬁle synchronization involves both uploading and

downloading ﬁles, it is important to not bias the opti-

mization in favor of one phase over the other. Equa-

tion 3 corresponds to uploading a ﬁle using the syn-

chronization algorithm, and later downloading the

same ﬁle back to the same device under the same net-

work conditions. This procedure is illustrated in Fig-

ure 2. In this test, a ﬁlesize of F = 1 MB was chosen.

The result of a distributed ﬁle synchronization al-

gorithm minimization is a set of conﬁguration param-

eters for that algorithm. In our chosen algorithm e.g.

Syncrod, the parameters are B, H, C, and N (see Sec-

tion 3.3 for explanations). C and N have a fairly lim-

ited set size (10 and 2 respectively), and multiple time

equations (t

com

, t

rec

, and t

dec

) depend solely upon C,

Figure 2: Overview of test procedure used to measure en-

ergy usage.

N, or the derived parameter R(C). Therefore, we de-

cided to enumerate all possible input parameter com-

binations of C and N and generate solutions for t

com

rec

, and t

dec

. The results are shown in Table 1.

Table 1: Empirical values determined for t

com

, t

rec

, and t

dec

for varying levels of compression C and N.

C N t

com

rec

dec

0 Wi-Fi 0.000000 0.068900 0.000000

0 3G 0.000000 0.068900 0.000000

1 Wi-Fi 0.219298 0.068350 0.111936

1 3G 0.219298 0.068350 0.111936

2 Wi-Fi 0.175132 0.068343 0.103453

2 3G 0.175132 0.068343 0.103453

3 Wi-Fi 0.179364 0.068336 0.105835

4 3G 0.200769 0.068334 0.104855

4 Wi-Fi 0.200769 0.068334 0.104855

5 3G 0.201235 0.068327 0.114478

5 Wi-Fi 0.201235 0.068327 0.114478

6 3G 0.210243 0.068324 0.113710

6 Wi-Fi 0.210243 0.068324 0.113710

7 3G 0.214303 0.068322 0.122977

7 Wi-Fi 0.214303 0.068322 0.122977

8 3G 0.207665 0.068321 0.120345

8 Wi-Fi 0.207665 0.068321 0.120345

9 3G 0.200630 0.068321 0.123269

9 Wi-Fi 0.200630 0.068321 0.123269

Given that we enumerated all possible combina-

tions of C and N, each of the rows in Table 1 pro-

vides a partial solution to the optimization. However,

for each partial solution, the remaining parameters

B and H must also be determined. Each solution is

an instance of the objective Equation 1, which con-

sists of a constrained nonlinear multivariable func-

tion. Matlab’s

fmincon

program was used to min-

imize the value of the objective function across the

range of possible values for B and H.

SYNCROD: AN ENERGY-MINIMIZATION METHOD FOR DISTRIBUTED FILE SYNCHRONIZATION

ALGORITHMS ON MOBILE DEVICES

139

5 RELATED WORK

File synchronization, ﬁle compression and mobile

computing are prevalent areas of research that are in-

creasingly overlapping in problem domain space be-

cause smartphones, with limited storage capacity, are

being used as data storage for documents, songs and

other ﬁle types. We categorize the related work of

this problem domain into two types: (1) distributed

ﬁle systems, and (2) ﬁle synchronization algorithms.

Distributed File Systems. Cloud distributed

ﬁle systems offer an alternative to peer-to-peer net-

works (Aymerich et al., 2008; Popovic and Hocen-

ski, 2010; Uppoor et al., 2010). Recent research on

the Cloud and smartphones show that ﬁle updating

for software-as-a-service applications is an efﬁcient

way to synchronize software on all devices that house

the program (Cusumano, 2010). Our work is com-

plementary to this approach, in that our approach can

be used to optimize the parameters of distributed ﬁle

system synchronization algorithms that can be ﬁt into

our formal model in Equation 2.

Data Synchronization Algorithms. As smart-

phones move in location, the devices will experience

ﬂuctuations in network signal strength (Fukushima

et al., 2005). Fukushima et al. discuss an algorithm

for scheduling automatic ﬁle synchronization based

upon the received signal strength indication of the net-

work. Speciﬁcally, smaller ﬁles synchronize in areas

of lower bandwidth and vice versa for larger ﬁles. Our

approach is complementary to this work in that it pro-

vides an alternate method of minimizing energy con-

sumption. We also compare total energy consumption

over Wi-Fi and 3G, with results found in Section 4.

Yan et al. present a low-latency algorithm that ﬁrst

compartmentalizes ﬁles into ﬁxed-byte blocks, then

hashes each block for fast equality comparison when

synchronization is done (Yan et al., 2008). The low-

latency algorithm is very similar to our approach, ex-

cept for the fact that Yan et al. focused on minimiz-

ing network trafﬁc while we worked to reduce energy

consumption.

6 CONCLUSIONS

As mobile devices have increased in popularity, it is

relatively common for users to have multiple comput-

ing devices, such as a desktop computer, a laptop, and

a smartphone, creating the need for distributed ﬁle

synchronization. The existing approaches for build-

ing distributed ﬁle synchronization algorithms were

developed for desktop platforms, where battery ca-

pacity is not a concern. Due to the high importance

of energy consumption on mobile devices, there is a

need for ﬁle synchronization algorithms that attempt

to minimize total energy consumption.

This paper presented a method of deriving the

distributed ﬁle synchronization algorithm parameters

that would result in the lowest overall energy usage.

In order to do this, Section 3 derived a formal model

for energy usage of a distributed ﬁle synchronization

algorithm. This formal model was then used in Sec-

tion 4 to derive the most energy-efﬁcient distributed

ﬁle synchronization algorithm parameter values. The

approaches presented in this paper can easily be ex-

tended to cover other speciﬁc distributed ﬁle synchro-

nization algorithms.

From our research on this topic, we learned the

following important lessons:

1. Network Choice is the Single Most Important

Decision to when Attempting to Minimize En-

ergy Usage. Using a 3G network results in a 10x

increase in both time and energy over a Wi-Fi net-

work.

2. Ratio of File Compression is the Second-

most Important Design Consideration when

Attempting to Minimize Energy Usage. It is

clearly worth examining ﬁle extensions to see if

the ﬁles are likely to experience no beneﬁts from

being minimally compressed (such as synchro-

nization of ﬁles already stored in a compressed

format) and avoiding compression for those ﬁles.

3. Dynamic Block Sizing Should be Investigated.

The optimum block size is different for 3G and

Wi-Fi. This encourages research into storing mul-

tiple versions of the block-hash table and using

the version most appropriate for the context of the

mobile device. As mobile storage increases, this

approach of sacriﬁcing data to gain energy could

provide a notable improvement over the current

traditional methodology of consistently using one

block size.

REFERENCES

Agarral, S., Starobinski, D., and Trachtenberg, A. (2002).

On the scalability of data synchronization protocols

for pdas and mobile devices. IEEE Network, pages

22–28.

Aymerich, F., Fenu, G., and Surcis, S. (2008). An ap-

proach to a cloud computing network. In Applica-

tions of Digital Information and Web Technologies,

2008. ICADIWT 2008. First International Conference

on the, pages 113 –118.

Balasubramanian, N., Balasubramanian, A., and Venkatara-

mani, A. (2009). Energy Consumption in Mobile

PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems

140

Phones: A Measurement Study and Implications for

Network Applications. In ICM ’09. ACM.

Cusumano, M. (2010). Cloud computing and saas as new

computing platforms. Commun. ACM, 53:27–29.

Fukushima, T., Takahashi, E., Narazaki, H., and Tamura, N.

(2005). A wireless agent for autonomous ﬁle synchro-

nization in a mobile environment.

Gartner (2010). Gartner says worldwide mobile

phone sales grew 35 percent in third quarter

2010; smartphone sales increased 96 percent.

http://www.gartner.com/it/page.jsp?id=1466313.

Lagerspetz, E., Tarkoma, S., and Lindholm, T. (2010).

Dessy: Search and synchronization on the move. Mo-

bile Data Management, pages 215–217.

Mudge, T. (2001). Power: A ﬁrst-class architectural design

constraint. Computer, 34(4):52–58.

Mummert, L. B., Ebling, M. R., and Satyanarayanan, M.

(1999). Exploiting weak connectivity for mobile ﬁle

access. Mobility, pages 306–321.

Popovic, K. and Hocenski, Z. (2010). Cloud computing

security issues and challenges. In MIPRO, 2010 Pro-

ceedings of the 33rd International Convention, pages

344 –349.

Shye, A., Scholbrock, B., Memik, G., and Dinda, P. (2010).

Characterizing and modeling user activity on smart-

phones: summary. Proceedings of the ACM SIGMET-

RICS conference on Measurement and Modeling of

Computer Systems.

Uppoor, S., Flouris, M. D., and Bilas, A. (2010). Cloud-

based Synchronization of Distributed File System Hi-

erarchies.

Yan, H., Irmak, U., and Suel, T. (2008). Algorithms for

Low-Latency Remote File Synchronization. In IEEE

INFOCOM 2008. IEEE.

SYNCROD: AN ENERGY-MINIMIZATION METHOD FOR DISTRIBUTED FILE SYNCHRONIZATION

ALGORITHMS ON MOBILE DEVICES

141