This page is a dump of notes on disk benchmarking in Linux, and is in great need of editing! Use at own peril.
Buffered disk reads (bypasses cache)
This method reads data sequentially from the disk, disregarding file boundaries.
[email protected]:~$ sudo hdparm -t /dev/sdb /dev/sdb: Timing buffered disk reads: 228 MB in 3.02 seconds = 75.52 MB/sec
[email protected]:~$ sudo hdparm -T /dev/sdb /dev/sdb: Timing cached reads: 3158 MB in 2.00 seconds = 1579.16 MB/sec
[email protected]:~$ dd if=/dev/zero of=testfile bs=8k count=100k; rm -f testfile 102400+0 records in 102400+0 records out 838860800 bytes (839 MB) copied, 12.7363 s, 65.9 MB/s
Writes and Cache-out
[email protected]:~$ time sh -c "dd if=/dev/zero of=testfile bs=8k count=100k; sync" ; rm -f testfile 102400+0 records in 102400+0 records out 838860800 bytes (839 MB) copied, 12.6832 s, 66.1 MB/s real 0m19.304s user 0m0.020s sys 0m1.232s
- Go to [http://www.iozone.org/], get link of latest build.
- wget http://www.iozone.org/src/current/iozone-3-420.i386.rpm
- yum install iozone-3-420.i386.rpm
If you also need to create graphs from the results of
iozone, you can additionally install
yum install gnuplot
bonnie++ benchmark is available on EPEL repository for CentOS. Note
that it will put load on the system on which it runs, so it's better run during less productive times.
[[email protected] ~]# bonnie++ -u root
Bonnie++ will output its data twice, once in an ASCII formatted table, and once more in a single line CSV. Echo the contents of the CSV line through bon_csv2html to get a more readable HTML table.