Está en la página 1de 7

World Academy of Science, Engineering and Technology 27 2007

Benchmarking of PVM and LAM/MPI Using


OSCAR, Rocks and Knoppix Clustering Tools
Muhammad Sharif, and Aman Ullah Khan

As we know that MD5 hash contains hexadecimal digits (0-f),


Abstract—Parallel and distributed computing plays an important so 16 files are required corresponding for hexadecimal digits.
role by dividing a big process into many small processes running As shown in Fig. 1:
parallel with the help of number of processors. The communication in
distributed and parallel processors takes place with the help of
different API’s. In this paper the performance of two API’s i.e.
PVM and LAM/MPI and three Clustering Tools OSCAR, Rocks and
Knoppix is analyzed, So that the best results could be achieved by the
right selection and combination of both API and Clustering Tool.

Keywords—PVM, LAM/MPI, Clustering Tools, OSCAR, Rocks,


Knoppix.

I. INTRODUCTION

P VM and MPI are the widely used Parallel API’s to


communicate in the distributed and parallel environments.
The performance of API does vary on different clustering
tools. The performance of PVM and LAM/MPI with
clustering tools OSCAR, NPACI Rocks and Parallel Knoppix
is analyzed. The application of “Attack on MD5 Hashed
Passwords” is used in the three above mentioned Linux based
clusters and the two API’s.

II. THE APPLICATION USED FOR ANALYSIS


The application used is dictionary attack [2] to break MD5
[3] hashes in a shortest possible time. The application is Fig. 1 The file structure
further divided in three sub applications.
For example, the hash value
1. Dictionary Creation 76b0b0b60246f4e8e92a39ef7dbea194, have the first character
2. Sorting 7, so this hash will store in the file named 7.txt. Similarly the
3. Searching hash f14e837db317a2bba9826144e7831b51 with first
character ‘f’ will be stored in the file named f.txt and so on.
For this dictionary of all possible combinations of MD5 Hashes are searched according to their first character from the
hashes is created first and then the dictionary is sorted using respective hash file and the searching will reduce by 16 times.
quick sort, and at last multiple hashes are searched from it. The cracking time is further improved by distributing the
Numbers of strategies are used to increase the performance of dictionary over the cluster. Hashes are searched parallel by
the application. i.e, each node over the cluster.
For fast searching Binary search is used. Binary search can
• Hash File Structure help to reduce the searching time. To apply this searching
• Improvement of Cracking Time by Faster Searching technique all the files containing hashes needed to be sorted
Techniques first on each node. Binary search is preferred over sequential
search, although sorting is an overhead in binary search but it
The each record of the hash file contains two entries i.e. the is one time process and searching will be performed several
MD5 hash and the corresponding password. The dictionary is times as needed, so the Quick Sort is used to sort the hashes in
divided into 16 files according to first character of the hash. all files.

Authors are with Department of Computer Science, COMSATS Institute of


Information Technology, Wah Cantt, Punjab, Pakistan (e-mail:
muhammadsharifmalik@yahoo.com, auk_pk@yahoo.com).

250
World Academy of Science, Engineering and Technology 27 2007

III. THE COMPARISON OF API’S USING DIFFERENT CLUSTER All results are gathered by using four characters dictionary
TOOLS on two and four node clusters with character set A-Z and 0-9.
To compare two API’s PVM (Parallel Virtual Machine) and The hardware used for this purpose is:
LAM/MPI (Local Area Multi-computer /Message Passing Processor: Pentium 4 processors (2 GHz) on each node
Interface, the application is divided into three Phases i.e. Ram: 256MB on master node and 128MB on slave nodes
HDD: 40 GB on each node.
The Dictionary Creation Phase Application: It creates the
dictionary of specified length. It is one time process.
The Sorting Phase Application: It sorts all files related to A. OSCAR
dictionary. It is also one time process. OSCAR [4] is an open source cluster tool developed by
The Searching Phase Application: It is used to search the open cluster group. It is software that installed on different
multiple hashes from the dictionary files as many times as Linux based operating systems. It is not related to an OS. It is
needed. easy to downloads, install and configure. It consists of easy to
install and fully integrated bundle of software designed for
The three sub applications (dictionary creation, sorting and high performance cluster computing.
searching) are implemented using the two API’s named PVM
and LAM/MPI with three cluster tools (OSCAR, Rocks and From the table and figures below it is observed that in
Parallel Knoppix) to analyze which API is better with which OSCAR Cluster tool LAM/MPI works better as it finds hashes
cluster tool. For this multiple hashes of MD5 were searched in less time than that of PVM. So LAM/MPI is a better choice
and their searching time is calculated. while using with OSCAR.

TABLE I
OSCAR RESULTS FOR FOUR CHARACTERS DICTIONARY USING 2 AND 4 NODES
PVM LAM/MPI
4Char, 2Nodes 4Char,4Nodes 4Char, 2Nodes 4Char, 4Nodes
Dictionary Creation Time (sec) 11626.96895 6547.233515 11597.54207 6045.257304
Sorting Time (sec) 2161.382428 1426.19975 2124.546293 1365.344605
Hashes
5 0.085192 0.01799 0.103596 0.011243
10 0.193572 0.160329 0.176136 0.087434
15 0.313313 0.277333 0.201219 0.154762
20 0.68193 0.31612 0.223072 0.182779
Searching
Time (sec) 25 0.926807 0.655534 0.251057 0.248653
30 1.252807 0.810098 0.292722 0.289317
35 1.63352 1.066407 0.333038 0.315672
40 2.071061 1.241928 0.851814 0.772133
45 2.507061 1.492638 1.528491 1.334176
50 2.982934 1.848599 1.901674 1.796431

OSCAR Dictionary Creation Time(Sec) OSCAR Sorting Time(Sec)

4 Charracters, 4 Charracters,
4 Nodes 4 Nodes 4 Characters,
4 Characters, LAM/MPI,
LAM/MPI, 2 Nodes
2 Nodes
6045.257304, 1365.344605, PVM,
PVM,
17% 19% 2161.382428,
11626.96895,
31%
33%

4 Characters, 4 Characters,
2 Nodes 2 Nodes
4 Charracters, 4 Charracters,
LAM/MPI, LAM/MPI,
4 Nodes 4 Nodes
11597.54207, 2124.546293,
PVM, PVM,
32%
6547.233515, 30% 1426.19975,
18% 20%

Fig. 2 OSCAR Dictionary Creation Results Fig. 3 OSCAR Sorting Results

251
World Academy of Science, Engineering and Technology 27 2007

B. NPACI Rocks
OSCAR Searching Application
NPACI Rocks [5] distribution (hereinafter Rocks) was
3.5
developed at San Diego Supercomputing Centre. Along with
3 OSCAR, Rocks is said to be the most commonly used cluster
2.5
distribution. The installation suit of Rocks is in the form of
Rolls such as Base Roll, Kernel Roll and HPC rolls etc. One
T im e (S e c )

4 Characters, 2 Nodes
2 PVM disadvantage of Rocks is that it is related to operating system.
1.5 4 Charracters, 4 Nodes The Red Hat Kick start mechanism is used to install client
PVM
1 4 Characters, 2 Nodes nodes also called compute nodes.
LAM/MPI By analyzing the results using NPACI Rocks cluster tool, it
0.5 4 Charracters, 4 Nodes
LAM/MPI
is observed that LAM/MPI works better than PVM like it is
0 efficient than PVM in OSCAR.
5 10 15 20 25 30 35 40 45 50
Hashes

Fig. 4 OSCAR Searching Results

TABLE II
ROCKS RESULTS FOR FOUR CHARACTERS DICTIONARY USING 2 AND 4 NODES
PVM LAM/MPI
4Char, 2Nodes 4Char,4Nodes 4Char, 2Nodes 4Char, 4Nodes
Dictionary creation Time (sec)
13009.25411 5410.262474 9259.218346 5377.230585
Sorting Time (sec)
2289.685124 1332.78622 2097.126034 1305.597563
Hashes
5 0.041937 0.040106 0.055235 0.05926
10 0.044651 0.054146 0.069312 0.109876
15 0.066918 0.099698 0.085959 0.133579
20 0.104115 0.328301 0.114163 0.14077
Searching
25 0.273197 0.434129 0.147085 0.211389
Time (sec)
30 0.64441 0.843124 0.456817 0.131953
35 1.225285 1.048385 1.035961 0.227971
40 1.676079 1.090299 1.196487 0.301945
45 2.097876 1.153291 1.874351 1.14137
50 2.496298 1.160351 2.224394 1.15227

Rocks Dictionary Creation Time(Sec) Rocks Sorting Time(Sec)

4 Charracters, 4 Charracters,
4 Nodes 4 Nodes
4 Characters,
LAM/MPI, LAM/MPI,
4 Characters, 2 Nodes
5377.230585, 1305.597563,
2 Nodes PVM,
16% 19%
PVM, 2289.685124,
13009.25411, 32%
4 Characters, 40%
2 Nodes
4 Characters,
LAM/MPI,
4 Charracter, 2 Nodes
9259.218346, 4 Charracter,
4 Nodes LAM/MPI,
28% 4 Nodes
PVM, 2097.126034,
PVM,
5410.262474, 30%
1332.78622,
16%
19%

Fig. 5 Rocks Dictionary Creation Results Fig. 6 Rocks Sorting Results

252
World Academy of Science, Engineering and Technology 27 2007

Rocks Searching Application


C. Parallel Knoppix
3 Parallel Knoppix [6] is a bootable CD that allows creation
2.5
of a working Linux cluster on a network in few minutes.
4 Characters, 2 Nodes
2
PVM Using Parallel Knoppix it is observed that the performance
T im e (S e c )

4 Charracter, 4 Nodes
1.5
PVM of both LAM/MPI and PVM in searching application is nearly
4 Characters, 2 Nodes
LAM/MPI
same as searching time of both is nearly equal to each other.
1
4 Charracters, 4 Nodes Hence Parallel Knoppix is good for both API’s when
LAM/MPI
0.5 searching of hashes is used. For other two applications
0 LAM/MPI still produced better results.
5 10 15 20 25 30 35 40 45 50
Hashes

Fig. 7 Rocks Searching Results

TABLE III
KNOPPIX RESULTS FOR FOUR CHARACTERS DICTIONARY USING 2 AND 4 NODES
PVM LAM/MPI
4Char, 2Nodes 4Char, 4Nodes 4Char, 2Nodes 4Char, 4Nodes
Dictionary creation Time (sec) 7968 4392 7394.42798 4042.565133

Sorting Time (sec) 167.011033 87.985376 166.421677 84.254584


Hashes
5 0.011716 0.007638 0.010224 0.007302
10 0.0249 0.014752 0.019158 0.013489
15 0.03351 0.023064 0.027463 0.027452
20 0.04903 0.028243 0.04915 0.030749
Searching
25 0.056904 0.034663 0.051786 0.033484
Time (sec)
30 0.069333 0.048517 0.059235 0.047573
35 0.077598 0.053303 0.073666 0.053172
40 0.094824 0.062314 0.088843 0.061406
45 0.103139 0.067851 0.100367 0.0674
50 0.112371 0.082743 0.112496 0.070624

Knoppix Dictionary Creation Time (Sec) Knoppix Sorting Time(Sec)

4 Charracters, 4 Charracters,
4 Nodes 4 Nodes
LAM/MPI, LAM/MPI, 4 Characters,
4 Characters, 2 Nodes
4042.565133, 84.254584,
2 Nodes PVM,
17% 17%
PVM, 7968, 167.011033,
34% 33%

4 Characters, 4 Characters,
2 Nodes 2 Nodes
4 Charracters, LAM/MPI, 4 Charracters,
LAM/MPI,
4 Nodes 166.421677, 4 Nodes
7394.42798,
33% PVM,
31% PVM, 4392,
87.985376,
18% 17%

Fig. 8 Knoppix Dictionary Creation Results Fig. 9 Knoppix Sorting Results

253
World Academy of Science, Engineering and Technology 27 2007

IV. THE COMPARISON OF CLUSTERING TOOLS USING


Knoppix Searching Application
DIFFERENT API’S
0.12
By implementing MD5 attack using PVM and LAM/MPI
0.1 with three cluster tools OSCAR, NPACI Rocks and Parallel
4 Characters, 2 Nodes Knoppix; it is observed that which API is better with which
0.08 PVM
cluster tool. The following results illustrate this.
Tim e(S ec)

4 Charracters, 4 Nodes
PVM
0.06
4 Characters, 2 Nodes
LAM/MPI A. PVM
0.04 4 Charracters, 4 Nodes By implementing application (the three phases) using PVM,
LAM/MPI
0.02 it is observed that, PVM shows better results by using with
Parallel Knoppix. NPACI Rocks comes at second and OSCAR
0 at the third place. From the table and figures below, it can be
5 10 15 20 25 30 35 40 45 50
Hashes
easily analyzed that PVM with Parallel Knoppix provides
excellent results than that of the other two clustering tools.
Fig. 10 Knoppix Searching Results

TABLE IV
PVM RESULTS
OSCAR Rocks Knoppix
4Char, 4Char, 4Char, 4Char, 4Char, 4Char,
2Nodes 4Nodes 2Nodes 4Nodes 2Nodes 4Nodes
Dictionary creation
11626.96895 6547.233515 13009.25 5410.262 7968 4392
time (sec)

Sorting Time (sec) 2161.382428 1426.19975 2289.685 1332.786 167.011033 87.98538

Hashes
5 0.085192 0.01799 0.041937 0.040106 0.011716 0.007638
10 0.193572 0.160329 0.044651 0.054146 0.0249 0.014752
15 0.313313 0.277333 0.066918 0.099698 0.03351 0.023064

Searching 20 0.68193 0.31612 0.104115 0.328301 0.04903 0.028243


Time 25 0.926807 0.655534 0.273197 0.434129 0.056904 0.034663
(sec)
30 1.252807 0.810098 0.64441 0.843124 0.069333 0.048517
35 1.63352 1.066407 1.225285 1.048385 0.077598 0.053303
40 2.071061 1.241928 1.676079 1.090299 0.094824 0.062314
45 2.507061 1.492638 2.097876 1.153291 0.103139 0.067851
50 2.982934 1.848599 2.496298 1.160351 0.112371 0.082743

PVM Dictionary Creation Time(Sec) PVM Sorting Time(Sec)

4Charracters,
4 Nodes
Rocks, 4 Characters, 4Charracters,
5410.262474, 2 Nodes 4 Nodes 4 Characters,
11% OSCAR, Rocks, 2 Nodes
11626.96895, 1332.78622, OSCAR,
4 Characters, 24% 18% 2161.382428,
2 Nodes 29%
Rocks,
13009.25411, 4 Charracters, 4 Characters,
27% 4 Nodes 2 Nodes 4 Charracters,
OSCAR, Rocks, 4 Characters, 2
4 Charracters, 4 Nodes
6547.233515, 2289.685124, Nodes
4 Nodes 4 Characters, Knoppix,
OSCAR,
13% 31%
Knoppix , 2 Nodes 167.011033, 1426.19975,
4 Charracters, 19%
4392, 9% Knoppix, 7968, 2%
4 Nodes
16% Knoppix,
87.985376, 1%

Fig. 10 PVM Dictionary Creation Results Fig. 11 PVM Sorting Results

254
World Academy of Science, Engineering and Technology 27 2007

PVM Searching Application


B. LAM/MPI
3.5
The LAM/MPI performed best with Parallel Knoppix than
3
4 Characters, 2 Nodes OSCAR and Rocks. Rocks and OSCAR comes at second and
OSCAR
4 Charracters, 4 Nodes third respectively. From the results mentioned below and
2.5 OSCAR
above, it can easily be judged that Parallel Knoppix is the best
Time(Sec)

4 Characters, 2 Nodes
2
Knoppix choice for both LAM/MPI and PVM API’s.
1.5 4 Charracters, 4 Nodes
Knoppix
1 4 Characters, 2 Nodes
Rocks
0.5 4Charracters, 4 Nodes
Rocks
0
5 10 15 20 25 30 35 40 45 50
Hashes

Fig. 12 PVM Searching Results

TABLE V
LAM/MPI RESULTS
OSCAR Rocks Knoppix
4Char, 4Char, 4Char, 4Char, 4Char, 4Char,
2Nodes 4Nodes 2Nodes 4Nodes 2Nodes 4Nodes
Dictionary creation
time (sec) 11597.54207 6045.257304 9259.218 5377.231 7394.428 4042.565
Sorting Time (sec)
2124.546293 1365.344605 2097.126 1305.598 166.4217 84.25458
Hashes
5 0.103596 0.011243 0.055235 0.05926 0.010224 0.007302
10 0.176136 0.087434 0.069312 0.109876 0.019158 0.013489
15 0.201219 0.154762 0.085959 0.133579 0.027463 0.27452
Searching 20 0.223072 0.182779 0.114163 0.14077 0.04915 0.030749
Time 25 0.251057 0.248653 0.147085 0.211389 0.051786 0.033484
(sec)
30 0.292722 0.289317 0.456817 0.131953 0.059235 0.047573
35 0.333038 0.315672 1.035961 0.227971 0.073666 0.053172
40 0.851814 0.772133 1.196487 0.301945 0.088843 0.061406
45 1.528491 1.334176 1.874351 1.14137 0.100367 0.0674
50 1.901674 1.796431 2.224394 1.15227 0.112496 0.070624

LAM/MPI Dictionary Creation Time LAM/MPI Sorting Time(Sec)

4Charracter
4 Characters, 4Charracter
4 Nodes
2 Nodes 4 Nodes 4 Characters,
Rocks,
OSCAR, Rocks, 2 Nodes
5377.230585, 1305.597563, OSCAR,
12% 11597.54207,
4 Characters, 18% 2124.546293,
27%
2 Nodes 31%
Rocks, 4 Characters,
9259.218346, 2 Nodes
21% Rocks,
4Charracters,
2097.126034,
4 Nodes 4 Character 4Charracters,
4Charracters, 29%
4 Character OSCAR, 2 Nodes 4 Nodes
4 Nodes 6045.257304, OSCAR,
2 Nodes Knoppix,
Knoppix, 14% 4Charracters, 166.421677, 1365.344605,
Knoppix,
4042.565133, 4 Nodes 2% 19%
7394.42798,
9% Knoppix,
17%
84.254584, 1%

Fig. 13 LAM/MPI Dictionary Creation Results Fig. 14 LAM/MPI Sorting Results

255
World Academy of Science, Engineering and Technology 27 2007

LAM/MPI Searching Application

2.5
4 Characters, 2 Nodes
OSCAR
2 4Charracters, 4 Nodes
OSCAR
Time(Sec)

1.5 4 Character 2 Nodes


Knoppix
4Charracters, 4 Nodes
1 Knoppix
4 Characters, 2 Nodes
0.5 Rocks
4Charracter 4 Nodes
Rocks
0
5 10 15 20 25 30 35 40 45 50
Hashes

Fig. 15 LAM/MPI Searching Results

V. CONCLUSION
Application contain three different phases i.e. Dictionary
Creation, Sorting and Searching using the two API’s PVM
(Parallel Virtual Machine) and LAM/MPI (Local Area Multi-
computer /Message Passing Interface) with the help of three
different Clustering Tools i.e. OSCAR(Open Source Cluster
Application Resource), Parallel Knoppix and NPACI Rocks.
Analysis from the above results proves that LAM/MPI with
Parallel Knoppix can provide better solution in most of the
situations.

REFERENCES
[1] PVM and MPI are completely different by William Gropp and Ewing
Lusk, Mathematics and Computer Science Division Argonne National
laboratory.
[2] Dictionary Attack, http://en.wikipedia.org/wiki/Dictionary_attack
[3] MD5 Homepage (unofficial),
http://userpages.umbc.edu/~mabzug1/cs/md5/md5.html
[4] OSCAR, http://oscar.openclustergroup.org/
[5] NPACI Rocks, http://www.rocksclusters.org/
[6] ParallelKnoppix, http://idea.uab.es/mcreel/ParallelKnoppix/

256

También podría gustarte