iRepository

International Conference on Information and

Communication Technologies

Aug 27th, 5:30 PM - 5:55 PM

A Performance Comparison of Data Encryption Algorithms

Aamer Nadeem
National University of Science and Technology, Rawalpindi, Pakistan

Dr. M. Younus Javed

National University of Science and Technology, Rawalpindi, Pakistan

Follow this and additional works at: https://ir.iba.edu.pk/icict

Part of the Databases and Information Systems Commons, and the Theory and Algorithms Commons

Recommended Citation
Nadeem, A., & Javed, D. Y. (2005). A Performance Comparison of Data Encryption Algorithms.
International Conference on Information and Communication Technologies. Retrieved from
https://ir.iba.edu.pk/icict/2005/2005/33

This document is brought to you by iRepository. For more information, please contact [email protected].
https://ir.iba.edu.pk/icict/2005/2005/33

A Performance Comparison of Data

Encryption Algorithms
Aamer Nadeem, Dr M. Younus Javed
Department of Computer Engineering, College of Electrical and Mechanical Engineering,
National University of Sciences and Technology,
Rawalpindi, Pakistan.
aamernadeemgyahoo.com, myjavedgceme.edu.pk

Abstract-The principal goal guiding the design of

any encryption algorithm must be security against In this paper, the security issues have been
unauthorized attacks. However, for all practical completely ignored, and only the performance of
applications, performance and the cost of the leading secret key algorithms has been
implementation are also important concerns. A compared on different platforms, using input
data encryption algorithm would not be of much data files of varying sizes and formats.
use if it is secure enough but slow in performance
because it is a common practice to embed
encryption algorithms in other applications such as While comparing the performance of algorithms,
e-commerce, banking, and online transaction the time required to set up the key(s) has been
processing applications. Embedding of encryption ignored.
algorithms in other applications also precludes a
hardware implementation, and is thus a major The rest of this paper is organized as follows:
cause of degraded overall performance of the section 2 gives a brief introduction of the
system. In this paper, the four of the popular secret algorithms that have been chosen for
key encryption algorithms, i.e., DES, 3DES, AES implementation; section 3 provides
(Rijndael), and the Blowfish have been
implemented, and their performance is compared implementation details; section 4 presents
by encrypting input files of varying contents and performance results; section 5 discusses the
sizes, on different Hardware platforms. The trade-offs between performance and security;
algorithms have been implemented in a uniform and finally section 6 concludes the work.
language, using their standard specifications, to
allow a fair comparison of execution speeds. The
performance results have been summarized and a II. IMPLEMENTED ALGORITHMS
conclusion has been presented. Based on the
experiments, it has been concluded that the The following secret key encryption algorithms
Blowfish is the best performing algorithm among
the algorithms chosen for implementation. were chosen for implementation, i.e.
* DES
Key Words: Data encryption, Cryptography * Triple DES
algorithms, Performance analysis. * AES (Rijndael)
* Blowfish
A. Data Encryption Standard (DES)
I. INTRODUCTION
DES (Data Encryption Standard) is currently the
The main consideration in designing an most widely used block cipher in the world. In
encryption algorithm has to be the security of the May 1973, NIST (then NBS) called for possible
algorithm against undesirable attacks. However, encryption algorithms for use in unclassified
in the real world, performance and adopted encryption algorithm and is in many
implementation cost is also important concerns. standards around the world (e.g. Australian
In this paper, security of the algorithms against Standard AS2805.5-1985). One of the largest
attacks has not been compared. The primary users of the DES is the banking industry. It is for
focus is on comparing the encryption algorithms this use that the DES was primarily standardized,
on the basis of their performance and ease of with ANSI reconfirming its use for 5 year
implementation. periods - in future it will be replaced with AES.

84
0-7803-9421-6/05/$20.00 ©2005 IEEE.
Published by iRepository, 2022
Authorized licensed use limited to: Trial User - Institute of Business Administration Karachi. Downloaded on May 10,2023 at 06:10:30 UTC from IEEE Xplore. Restrictions apply.
https://ir.iba.edu.pk/icict/2005/2005/33

Although the DES standard is public, the design 1. 128-bit block size
criteria used are classified. There has been 2. 128, 192, or 256 bit key size
considerable controversy over the design, 3. an iterative rather than a Feistel cipher
particularly in the choice of a 56-bit key [1]. (like IDEA)
4. treats data as 4 groups of 4 bytes
B. Triple DES (3DES) 5. has 9, 11, or 13 rounds, where each
round consists of:
The triple DES (3DES) algorithm was needed as * a byte substitution step (1 S-
a replacement for DES due to advances in key box used
searching. 3DES is a proposal based on the * on every byte)
existing DES, and was standardized in ANSI * a shift rows step (shuffle the
X9.17 & ISO 8732 and in PEM for key bytes
management. It was also proposed for general * between groups)
EFT standard by ANSI X9 [1]. It is backwards * a mix columns step (matrix
compatible with existing single DES (when * multiplication of groups with
K1=K2=K3). The 3DES algorithm uses either each
two or three 56-bit keys. Thus the effective key * other)
length is up to 168 bits. 3DES is defined by the * an add round key step
following function: 6. all operations can be combined into xor
C = DESK3 {DES-i and table lookups - hence
K2{DESKl(P)}
implementation can be very fast and
efficient.
where P = Plaintext D. Blowfish
C = Ciphertext
DESK= DES encryption using key K Most of the encryption algorithms today are
DES-i
K= DES decryption using key K unavailable to the public - many of them are
protected by patents (e.g. Khufu, REDOC II, and
C. Advanced Encryption Standard (AES) IDEA), or being kept secret by the governments
(e.g. Skipjack and Capstone are protected by the
In September 1997, US NIST announced a call U.S. government). Many of the other algorithms
for candidate ciphers for its new Advanced are available only in part (e.g. RC2, RC4, and
Encryption Standard (AES), because clearly a GOST). Bruce Schneier - one of the world's
replacement for DES was needed at that time [2]. leading cryptologists, and the president of
The candidate ciphers were to be submitted by Counterpane Systems, a consulting firm
June 1998, and a finalist was selected in October specializing in cryptography and computer
2000. In total 15 candidates were accepted in security - designed the Blowfish algorithm [5]
June 98 (6 were rejected as incomplete), and 5 and made it available in the public domain.
were short-listed in August 99. Finally, Rijndael Blowfish is a variable length key, 64-bit block
was selected as the AES finalist in October 2000. cipher. It was his intent from the outset of
creating this new encryption algorithm to
NIST has released all submissions and provide the world with a new encryption
unclassified analyses. The AES candidates are standard. The Blowfish algorithm was first
the latest generation of block ciphers, and have a introduce in 1993, and has not been cracked yet.
significant increase in the block size - from the It is also noteworthy to point out that this
old standard of 64-bits up to 128-bits; and keys algorithm can be optimized in hardware
from 128 to 256-bits. In part this has been driven applications, although it, like most other ciphers,
by the public demonstrations of exhaustive key is often used in software applications.
searches of DES & RC-5 (at 64-bits).
After exhaustive analysis and evaluation rounds, III. IMPLEMENTATION DETAILS
Rijndael, designed by Rijmen & Daemen in
Belgium [4], was selected as AES finalist [3]. It The Java platform (JDK 1.4) was used to
has the following attributes: implement the above algorithms. The following

85

Published by iRepository, 2022

Authorized licensed use limited to: Trial User - Institute of Business Administration Karachi. Downloaded on May 10,2023 at 06:10:30 UTC from IEEE Xplore. Restrictions apply.
https://ir.iba.edu.pk/icict/2005/2005/33
are some of the main reasons explaining why implementation of cryptography algorithms - but
Java was chosen: just to compare the relative performance of
various popular algorithms. So the algorithms
* Java is considered platform were implemented as is, using a uniform
independent because Java compiler language, and were tested on a uniform platform.
produces byte code rather than Each of the above algorithms was implemented
machine code for a specific type of as a Java class. In the subsequent sections, for
hardware - this feature of Java each algorithm, a description of the instance
makes sure that the programs will variables and the implemented methods is
run on any platform (with Java provided.
interpreter). Thus, the implemented
algorithms can be tested on a
variety of platforms for comparison IV. PERFORMANCE RESULTS
purposes.
* Java (in particular JDK 1.4) All the implementations provided and discussed
provides a large library of built-in here have been coded from scratch using the
classes and methods (in the form of specification documents of respective
API) that assist the programmer in algorithms.
writing code for cryptographic
algorithms. For example, the A. Performance Measurement Approach
BigInteger class in Java lets the
programmer apply arithmetic and A fairly consistent style across the source code
bit manipulation operations on was adopted to ensure a fair comparison. All the
integer values of arbitrarily large routines have been implemented as per
sizes. specifications without any bias towards
* Conversion from integer to string encryption and decryption speed, or to key set-up
and vice versa, and likewise or to memory use in tables. The main reason for
conversion of integer values from not optimizing the code for performance was to
one radix to another is relatively measure only relative performance of
easier in Java due to the built-in algorithms. Moreover, techniques for enhancing
routines provided for this purpose. performance of algorithms generally also make
For instance, one frequently needs the source code more obscure and unreadable.
to convert a decimal value to binary
or hexadecimal, and vice versa. All the algorithms and results reported are based
* The concepts of object on implementation using Java language (JDK
serialization and stream 1.4). In general, custom routines were written for
input/output make it easy to read helper methods such as left-shift operation,
and write objects to external disk binary to hex conversion and vice versa, xor
files operation on two bit strings etc. instead of using
library functions. However, whenever
However, using Java to implement convenient, Java's built-in class library was also
cryptography algorithms has some drawbacks as used, for instance the BigInteger class was used
well. The main drawback of using Java is its to manipulate integer values of arbitrarily large
slow speed - this is because Java compiler does size such as keys and data blocks.
not generate native machine code, rather it
produces an intermediate form code (called byte B. Measuring Execution Times
code) which needs an interpreter to run. This
could have been a concern because the Obtaining accurate and repeatable execution time
performance of various algorithms had to be measurements proved to be more difficult than
tested, but the effect of inefficiency was was originally anticipated. After many
balanced out because all the algorithms were experiments, it was decided to settle on the use
implemented in the same language (Java) and of a Pentium-I1 266 MHz machine (running
were tested on the same platform. Microsoft Windows operating system) and a
Pentium-4, 2.4 GHz machine (running Microsoft
As mentioned earlier, the primary goal of Windows XP operating system) as the basis for
this research was not to have the most efficient time measurements. Because the primary goal

86
Published by iRepository, 2022
Authorized licensed use limited to: Trial User - Institute of Business Administration Karachi. Downloaded on May 10,2023 at 06:10:30 UTC from IEEE Xplore. Restrictions apply.
https://ir.iba.edu.pk/icict/2005/2005/33

was to measure the encryption times of key algorithms are in the following order, as
algorithms, the initialization and key set-up regards their performance:
times were excluded from the comparison. Also, 1. Blowfish (fastest)
because decryption time is generally the same as 2. DES
encryption time for almost all the algorithms, 3. AES
therefore only the encryption times were 4. Triple DES (slowest)
measured.
In general, the performance of a block cipher
C. Performance Results for Block Ciphers varies with the block size and the key size. The
larger the block size, the faster will be the
The execution results of secret key algorithms in algorithm, because with a larger block size, a
ECB (Electronic Codebook) mode are presented large chunk of the data will be encrypted in a
first. These results are shown Table 1 and Table single execution cycle of the algorithm. With a
2, for execution of code on a Pentium-I1 266 smaller block size, the same size of input data
MHz machine, and Pentium-4, 2.4 GHz would require more execution cycles of the
machine respectively. algorithm, thus increasing the overall encryption
time other things being equal. Likewise, a larger
Table 1: Comparative execution times (in key results in a slower algorithm, because in
seconds) of secret key algorithms in ECB general, all bits of the key are involved in an
mode on a Pentium-l 266 MHz machine execution cycle of the algorithm. With a smaller
.i! DO 90ESi AS BF key, a lower number of key bits are involved,
thus reducing the time to complete one execution
20,W 24 72 39 19 cycle.
0w 48 123 74 35
. 1 5 ,
158 94
5982 74 202 126 53
69' 646
i371325 16
83 243
461
.143
235
6
136
D. Performance Results for Stream Ciphers
15.8959 19 543 324 158
i5 198 569 355.62 In this section, the results of execution of the
; 1 83- 227 655 378 176 secret key algorithms in CFB mode are
.232 276 4M6
79 219
Av.rape presented. Tables 3 and 4 give execution times of
time 134 383 228 108 these algorithms on Pentium-Il 266 MHz, and
Pentium-4, 2.4 GHz platforms respectively.
8yaiw1.35 20 491 ..

Table 3: Comparative execution times of

Table 2: Comparative execution times (in secret key algorithms im CFB mode on a
seconds) of secret key algonrthms in ECB Pentium-Il 266 MHz machine
mode on a Pentium-4, 2.4 GHz machine .witske Ws 3ES AES BF
jjumjsie Es 3DES AES BF 20,527 t88 496 598 141
36, 2 362 934 1123 " IF
l20,52 2 7 4 2 .45.911 431 1174 14.84 3
36X100 4 13 6 3
59,821 570 1532 1932 433
45.911 5 17 8 4
e4 623 1894 2251 7518
N;862,
9 7 23 11 6 5 1T20 3622 4419 9l 2
69r646 9 26 13 7
1!rg591405 4240 504 1175
13,5| 17 51 26 14 5501
158,959 20 60 30 16 165,64 71511 4424 1216
1 62 31 17 191,38 1714 5123 5623 1369
166,364 232,39 2139 6231 7231 1649
191,383 24 72 36 19 Avwag
23,9 30 87 44 24 tim 101 2,900 M 812-1
Avwrag
timn 14 42 21 11
IBIt 7,98B 2, 5.2 .1.67I
10
Bytnh4¢ t10 X8

As with block ciphers, the performance of a

31 138

An obvious way to compare the performance of stream cipher also varies with the block size and
these algorithms is to take average of execution the key size, but the effect of a larger block size
times for each algorithm, and rank them is reversed. The larger the block size, the slower
accordingly. Using this criterion, it is apparent will be the algorithm, because with a larger block
from Tables 1 and 2 that the four symmetric size, the algorithm will have to do more work for

87
Published by iRepository, 2022
Authorized licensed use limited to: Trial User - Institute of Business Administration Karachi. Downloaded on May 10,2023 at 06:10:30 UTC from IEEE Xplore. Restrictions apply.
https://ir.iba.edu.pk/icict/2005/2005/33

the same amount of input data (a bit or a byte) in rounds needed to make a brute force key search
a single execution cycle of the algorithm. With a the most efficient form of attack. This is a
smaller block size, the same size of input data controversial suggestion that some cryptologists
would be encrypted more efficiently, thus do not accept even in principle. Therefore, it is
decreasing the overall encryption time other proposed to look at this as follows:
things being equal.
* the principle of measuring performance
Table 4: Comparative execution times of of an algorithm by timing the minimum
secret key algorithms in CFB mode on a number of secure rounds for each
Pentium-4, 2.4 GHz machine algorithm is useful, but
. tuL DES DES AS BF * there is no easy way of obtaining
impartial and widely accepted values
.
36.002
17
30 9.
58 62.
94
1
23
for the minimum number of secure
911 41 130 1C25 33 rounds for each algorithm.
59,862 52 181 174 46
9,64 9 2 200 53 However as an AES algorithm contributor, Eli
Jt5 129 401 409 1
158,959 15 472 473 122 Biham is not seen as an impartial observer and it
166.354 159 488 489 t30 would hence be preferable to have a wider
191383 185 568 567 148 community basis for setting the values to be used
232t98 229, 681 587 184
in such a process.
Averag
time 105 328 328 Be
A. Implementation and Security Issues
BytaIec 1i 55 341 341 1300
There are a significant number of algorithms that
The effect of a larger key in a stream cipher is are considered to be secure at an algorithm level.
the same as in a block cipher, i.e. it slows down In practice, however, most failures in
the encryption, because in general, all bits of the cryptographic systems derive not from
key are involved in an execution cycle of the weaknesses in the algorithms used but rather
algorithm. With a smaller key, a lower number from the exploitation of subtle flaws in the way
of key bits are involved, thus reducing the time the algorithms are implemented or through the
to complete one execution cycle. exploitation of interactions between algorithm
implementations and the environments in which
they operate. This means that it is vital to judge
V. PERFORMANCE VS. SECURITY TRADE- implementations not simply from the perspective
OFFS of the performance that they provide but also
from the perspective of implementation
The work presented here is primarily about assurance i.e. can one be confident that the
performance but it does not seem sensible to implementation does what the algorithm
consider this in complete isolation from security requires, no more and no less, and can one also
since there are inevitably trade-offs between be confident that it does not interact with the
these attributes in the real world scenario. One wider environment in ways that can be exploited
obvious example is the number of rounds to undermine the security that the algorithm
employed by each algorithm - a large number of itself is otherwise capable of providing. To
rounds make the algorithm slower but are answer these questions much more hard evidence
supposed to provide greater security [6]. needs to be derived from the implementation of
the algorithms in environments that are
Likewise, the quality of each round in achieving representative of practical use. In particular,
the mixing that the cipher is intended to provide there is very little evidence at the moment of
is also a factor that affects performance of an how the algorithms behave in low-end
algorithm. Thus, there is always a trade-off environments of the kind that may apply in much
between security and performance in similar of electronic commerce and consumer
algorithms - such as symmetric key algorithms. applications.
Eli Biham [7] has suggested that performance of
algorithms should be measured by timing the There is also a need to understand whether the
minimum number of secure rounds for each different algorithms are architecturally different
algorithm - that is the estimated number of in their ability to support implementation

88
Published by iRepository, 2022
Authorized licensed use limited to: Trial User - Institute of Business Administration Karachi. Downloaded on May 10,2023 at 06:10:30 UTC from IEEE Xplore. Restrictions apply.
https://ir.iba.edu.pk/icict/2005/2005/33

assurance techniques. None of the existing REFERENCES

specifications are precise enough to support
formal design and analysis methods and it even [1] National Bureau of Standards - Data Encryption
seems possible that at least some of the Standard, FIPS Publication 46, 1977.
[2] NIST, "Advanced Encryption Standard Call", NIST,
algorithms on offer are sufficiently complex to 1997. http://www.nist.gov/aes/
make these techniques difficult or even [3] NIST Advanced Encryption Standard (AES)
impossible to apply. This may also be true of Development Effort web site
other approaches for implementation assurance. http://csrs.nist.gov/encryption/aes/aes-home.htm
[4] Daemen, J., Rijmen, V.: "AES Proposal: Rijndael",
This is an area where much more work is needed Banksys/Katholieke Universiteit Leuven, Belgium, AES
before any final algorithm choices are made. submission, Jun 1998.
[5] Schneier, B.: "Description of a New Variable-Length
Key, 64-Bit Block Cipher (Blowfish)", Fast Software
Encryption, Cambridge Security Workshop Proceedings
VI. CONCLUSION AND FUTURE WORK (Dec. 1993), Lecture Notes in Computer Science (LNCS),
Springer-Verlag, Vol. 809, pp. 191-204, 1993, ISBN 3-540-
In this paper, the popular secret key algorithms 58108-1.
including DES, 3DES, AES (Rijndael), [6] Schneier, B., Kelsey, J., Whiting, D., Wagner, D., Hall,
C., and Ferguson, N.: "Performance Comparison of the AES
Blowfish, were implemented, and their Submissions", Counterpane Systems, Dec 1998.
performance was compared by encrypting input http://www.counterpane.com/AESperformance.html
files of varying contents and sizes. The [7] Biham, E.: "Design Tradeoffs of the AES Candidates", in
algorithms were implemented in a uniform Asiacrypt'98, Lecture Notes in Computer Science, Springer-
Verlag, 1998.
language, using their standard specifications, and
were tested on two different hardware platforms,
to compare their performance. In the end, the
results were presented which conclude that the
Blowfish is the fastest algorithm. Though
security was not catered for, in practice,
however, one would consider the security first. A
proposed direction for the future work could be
to analyze the performance/security trade-off in
greater depth. For instance, an algorithm with
more complex rounds and a larger number of
rounds is generally considered more secure. The
impact of these and other such factors on the
overall performance of an algorithm needs to be
measured.

89
Published by iRepository, 2022
Authorized licensed use limited to: Trial User - Institute of Business Administration Karachi. Downloaded on May 10,2023 at 06:10:30 UTC from IEEE Xplore. Restrictions apply.