Parallel BZIP2 - Benchmark Results

Parallel BZIP2 (PBZIP2)

Benchmark Results

by Jeff Gilchrist

This page contains the benchmark results from running BZIP2 and PBZIP2 on various platforms using several sets of data. All experiments are measured as wall clock times in seconds. This includes the time taken to read from input files and write to output files. Each experiment was carried out three times and the average of the three results was recorded. Since bzip2 does not take advantage of multiple processors, the running time is the same no matter how many processors the system has.

Benchmark results for compressing 1.83GB of data on a SunFire 6800.

Benchmark results for compressing 1.83GB of data on various machines.

Athlon MP-2600+

The platform used in this experiment was an AMD Athlon-MP 2600+ machine with two 2.1 GHz processors, 256KB L2 cache, and 1 GB of RAM. The software was compiled with gcc v3.3.1 under cygwin 1.5.10-3 on Windows XP Pro.

Data: Source code for Linux 2.4.23 kernel (linux-2.4.23.tar): 166,604,800 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
29,832,609

100.862

pbzip2 -p1
29,897,521

98.564

1.00

pbzip2 -p2
29,897,521

55.872

1.76

88.2%

Data: Database of elliptic curve distinguished points from ECC2-109 project: 1,966,717,056 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
1,647,492,885

1625.519

pbzip2 -p1
1,647,522,559

1553.417

1.00

pbzip2 -p2
1,647,522,559

884.052

1.76

87.9%

Opteron 240

The platform used in this experiment was an AMD Opteron 240 machine with two 1.4 GHz processors, 1MB L2 cache, and 1 GB of RAM. The software was compiled with gcc v3.2 under Linux 2.4.25.

Data: Source code for Linux 2.4.23 kernel (linux-2.4.23.tar): 166,604,800 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
29,832,609

95.766

pbzip2 -p1
29,897,521

93.289

1.00

pbzip2 -p2
29,897,521

45.503

2.05

102.5%*

* - While super-linear speedup in the algorithm should not be possible, there must have been something hardware related that limited the speed of the system (disk IO perhaps) when running with 1 processor, allowing for a super-linear result with two processors.

Data: Database of elliptic curve distinguished points from ECC2-109 project: 1,966,717,056 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
1,647,492,885

1756.939

pbzip2 -p1
1,647,522,559

1618.409

1.00

pbzip2 -p2
1,647,522,559

868.250

1.86

93.2%

Pentium4 3.06 Xeon

The platform used in this experiment was an Intel Pentium4 Xeon machine with two 3.06 GHz processors (and two virtual hyperthreaded processors), 512KB L2 cache, and 2 GB of RAM. The software was compiled with gcc v3.2.2 under Linux 2.4.20-8smp.

Data: Source code for Linux 2.4.23 kernel (linux-2.4.23.tar): 166,604,800 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
29,832,609

64.477

pbzip2 -p1
29,897,521

63.127

1.00

pbzip2 -p2
29,897,521

37.155

1.70

85.0%

pbzip2 -p3
29,897,521

36.040

1.75

58.4%

pbzip2 -p4
29,897,521

36.353

1.74

43.4%

Data: Database of elliptic curve distinguished points from ECC2-109 project: 1,966,717,056 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
1,647,492,885

1414.985

pbzip2 -p1
1,647,522,559

1206.061

1.00

pbzip2 -p2
1,647,522,559

859.347

1.40

70.2%

pbzip2 -p3
1,647,522,559

727.156

1.66

55.3%

pbzip2 -p4
1,647,522,559

722.746

1.67

41.7%

Itanium2 900

The platform used in this experiment was an Intel Itanium2 machine with two 900 MHz processors, 1.5MB L3 cache, and 4 GB of RAM. The software was compiled with gcc v2.96 and run on RedHat Linux 7.2 with the 2.4.19-4 SMP kernel.

Data: Source code for Linux 2.4.23 kernel (linux-2.4.23.tar): 166,604,800 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
29,832,609

96.194

pbzip2 -p1
29,897,521

95.007

1.00

pbzip2 -p2
29,897,521

48.214

1.97

98.5%

Data: Database of elliptic curve distinguished points from ECC2-109 project: 1,966,717,056 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
1,647,492,885

3175.782

pbzip2 -p1
1,647,522,559

3061.973

1.00

pbzip2 -p2
1,647,522,559

1579.783

1.94

96.9%

SunFire 6800

The platform used in this experiment was a SunFire 6800 machine with twenty four 1.05 GHz UltraSPARC-III processors, 8MB L2 cache, and 96 GB of RAM. The software was compiled with gcc v3.2.3 and run on SunOS 5.9.

Data: Source code for Linux 2.4.23 kernel (linux-2.4.23.tar): 166,604,800 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
29,832,609

93.373

pbzip2 -p1
29,897,521

95.080

1.00

pbzip2 -p2
29,897,521

47.836

1.99

99.4%

pbzip2 -p4
29,897,521

24.419

3.89

97.3%

pbzip2 -p6
29,897,521

16.292

5.84

97.3%

pbzip2 -p8
29,897,521

12.408

7.66

95.8%

pbzip2 -p10
29,897,521

10.389

9.15

91.5%

pbzip2 -p12
29,897,521

8.551

11.12

92.7%

pbzip2 -p14
29,897,521

7.437

12.79

91.3%

pbzip2 -p16
29,897,521

6.601

14.40

90.0%

pbzip2 -p18
29,897,521

5.994

15.86

88.1%

pbzip2 -p20
29,897,521

5.537

17.17

85.9%

pbzip2 -p22
29,897,521

5.201

18.28

83.1%

Data: Database of elliptic curve distinguished points from ECC2-109 project: 1,966,717,056 bytes

Program

Size (bytes)

Time (sec)

Speedup

% of Linear Speedup

bzip2
1,647,492,885

1721.722

pbzip2 -p1
1,647,522,559

1644.484

1.00

pbzip2 -p2
1,647,522,559

830.029

1.98

99.1%

pbzip2 -p4
1,647,522,559

416.103

3.95

98.8%

pbzip2 -p6
1,647,522,559

278.090

5.91

98.6%

pbzip2 -p8
1,647,522,559

208.676

7.88

98.5%

pbzip2 -p10
1,647,522,559

167.235

9.83

98.3%

pbzip2 -p12
1,647,522,559

139.418

11.80

98.3%

pbzip2 -p14
1,647,522,559

119.839

13.72

98.0%

pbzip2 -p16
1,647,522,559

104.929

15.67

98.0%

pbzip2 -p18
1,647,522,559

93.088

17.67

98.1%

pbzip2 -p20
1,647,522,559

83.993

19.58

97.9%

pbzip2 -p22
1,647,522,559

76.736

21.43

97.4%

This web page is maintained by Jeff Gilchrist, Copyright © 2003-2008.

This web page best viewed using a resolution of 800 x 600 or higher.