tiobench benchmark
Introduction
Results
Test Environment
Resources
6 October 2002. JFS performance under kernel 2.5.40 is equal or better than EXT3 and ReiserFS. At the Linux World Conference and Expo in San Francisco, August 2002, Randy Dunlap has reported that JFS sequential write performance is very slow compared to other journaling file systems [Linux Journaling Filesystems and Workloads] when running multithreaded tiobench in an SMP environment. A problem was found to be in the JFS allocation code which might cause disk fragmentation when subject to 2 or more tiobench threads under SMP. The allocation patch for JFS was integrated into 2.5.40.
Test Specifications:
- 4-way SMP 500 MHz, 2.4 GB RAM, 2-9.2 GB hard drives.
- Linux kernel 2.5.40
- tiobench command: tiotest -f120 -t[numberofthreads] -b4096
- All journaling filesystems were selected and configured with menuconfig, and were created and mounted with default value.
The following data represent the average of three runs for each I/O test. The data are throughput rate in MB/s.
JFS EXT3 ReiserFS threads=1 write 11.04 10.51 9.52 random write 0.54 0.54 0.54 read 202.47 199.60 201.01 random read 184.22 184.24 186.08 threads=2 write 12.00 10.17 5.02 random write 0.56 0.56 0.57 read 278.43 283.26 283.66 random read 267.51 275.47 271.75 threads=4 write 12.22 7.44 5.38 random write 0.60 0.60 0.60 read 309.85 309.86 308.34 random read 319.33 320.76 318.96 threads=6 write 12.17 1.24 5.05 random write 0.44 0.62 0.62 read 310.32 310.50 308.25 random read 298.50 302.31 301.56
Last updated: 7 October 2002