Tech Every Where: Ext3 Vs. Ext4

Wednesday, August 5, 2009

Ext3 Vs. Ext4

We’ve started to do some internal benchmarking of ext3 vs ext4 at myYearbook.com to see if what we’ve seen and heard about ext4 was really true. While the following benchmark is not in-depth, it does represent our initial findings, which match our anecdotal findings. If all of these findings hold true, we expect them to have a large impact on our PostgreSQL OLTP workload where machines are IO bound.

The test platform:

Dell r905
Quad, Quad Core AMD Opteron(tm) Processor 8360 SE
128GB RAM
Red Hat Enterprise Linux Server release 5.3
2x Dell MD1120 DAS Arrays, 1 Perc 6/E per DAS
48 Seagate 2.5” SAS 15k RPM 72GB Drives
Each MD1120 has 2x RAID10 LUN’s
Linux kernel RAID0 across all 4 LUN’s

The Results:

EXT4DEV
"  Initial write "  214776.55
"        Rewrite "  305409.42
"           Read "  361373.46
"        Re-read " 9440588.47
"    Random read " 1452105.32
" Mixed workload " 1327560.92
"   Random write "  101430.37

EXT3
"  Initial write "  199546.18
"        Rewrite "  340091.72
"           Read "   91159.31
"        Re-read "   93897.52
"    Random read "   52234.47
" Mixed workload "  276443.58
"   Random write "   92115.62

Larger values are better. As you can see, in most cases ext4 is faster than ext3 and in the case of re-read, we initially thought the results were wrong and double checked the findings.