I'm on a Linode 360, and found a little benchmark script that benchmark's disk speed among other things:

Start: 2009-10-12 00:15:30
Create a 1GB file.
00:15:30 up 23:54, 1 user, load average: 0.18, 0.57, 0.62

1048576000 bytes (1.0 GB) copied, 217.374 seconds, 4.8 MB/s

00:19:08 up 23:58, 2 users, load average: 2.03, 1.36, 0.92
gzip of 1GB file
real 92.42
md5sum of gzip 1GB file
real 11.40
End: 2009-10-12 00:20:53


Only 4.8 MB/s? Why so low? This is a stock CentOS install that is not very active at all as I have not moved the sites over to it yet.

Any ideas what's going on?

[rjones@server2 ~]$ uname -a
Linux server2.xxxxxx.com 2.6.18.8-linode19 #1 SMP Mon Aug 17 22:19:18 UTC 2009 i686 i686 i386 GNU/Linux

IIRC there may be as many as 40 linodes on a 360 host, all sharing the same disks. I don't think 4.8M/s is unreasonable if you consider the probability that many other linodes are accessing the disk at the same time.

How exactly were you creating the file?

Just used dd.

Here's the script in its entirety:


#! /bin/sh
# Author: chriss
# Date: 2008/09/13
# Simple benchmarks

TFILE=./ddtmp
TFILEGZ=${TFILE}.gz
SRCFILE=/dev/urandom
NULL=/dev/null
GREPBIN=`which grep`
WBIN=`which w`
HEADBIN=`which head`
MD5SUMBIN=`which md5sum`
TAILBIN=`which tail`
GZIPBIN=`which gzip`
DDBIN=`which dd`
DATEBIN=`which date`
TEEBIN=`which tee`

# Quick sanity checks
if [ ! "$GREPBIN" ]; then
echo "Utility grep not found. Exiting."
exit 1
fi

if [ ! "$WBIN" ]; then
echo "Utility w not found. Exiting."
exit 1
fi

if [ ! "$HEADBIN" ]; then
echo "Utility head not found. Exiting."
exit 1
fi

if [ ! "$MD5SUMBIN" ]; then
echo "Utility md5sum not found. Exiting."
exit 1
fi

if [ ! "$TAILBIN" ]; then
echo "Utility tail not found. Exiting."
exit 1
fi

if [ ! "$GZIPBIN" ]; then
echo "Utility gzip not found. Exiting."
exit 1
fi

if [ ! "$DDBIN" ]; then
echo "Utility dd not found. Exiting."
exit 1
fi

if [ ! "$DATEBIN" ]; then
echo "Utility date not found. Exiting."
exit 1
fi

if [ ! "$TEEBIN" ]; then
echo "Utility tee not found. Exiting."
exit 1
fi

# Do a simple benchmark
LOGFILE="./${0}.`$DATEBIN +%Y%m%d%H%M%S`.log"
echo "Start: `$DATEBIN "+%Y-%m-%d %H:%M:%S"`" | $TEEBIN -a $LOGFILE
echo "Create a 1GB file." | $TEEBIN -a $LOGFILE
$WBIN | $HEADBIN -1 | $TEEBIN -a $LOGFILE
$DDBIN if=$SRCFILE of=$TFILE bs=1024 count=1024000 2>&1 | $TAILBIN -1 | $TEEBIN -a $LOGFILE
$WBIN | $HEADBIN -1 | $TEEBIN -a $LOGFILE
echo "gzip of 1GB file" | $TEEBIN -a $LOGFILE
{ time -p $GZIPBIN $TFILE 1>$NULL 2>&1 ; } 2>&1 | $GREPBIN real | $TEEBIN -a $LOGFILE
echo "md5sum of gzip 1GB file" | $TEEBIN -a $LOGFILE
{ time -p $MD5SUMBIN $TFILEGZ 1>$NULL 2>&1 ; } 2>&1 | $GREPBIN real | $TEEBIN -a $LOGFILE

rm $TFILEGZ

echo "End: `$DATEBIN "+%Y-%m-%d %H:%M:%S"`" | $TEEBIN -a $LOGFILE

I tried your script on two Linodes, a 360 and a 540.

360

Start: 2009-10-12 19:36:35
Create a 1GB file.
 19:36:35 up 98 days, 23:55,  3 users,  load average: 0.00, 0.05, 0.09
1048576000 bytes (1.0 GB) copied, 219.067 s, 4.8 MB/s
 19:40:14 up 98 days, 23:59,  3 users,  load average: 1.54, 0.69, 0.33

540

Start: 2009-10-12 18:00:24
Create a 1GB file.
 18:00:24 up 5 min,  2 users,  load average: 0.00, 0.00, 0.00
1048576000 bytes (1.0 GB) copied, 145.728 s, 7.2 MB/s
 18:02:50 up 7 min,  2 users,  load average: 1.47, 0.64, 0.24

Looks as if Stever may be right - in this rather limited test, the rate of disk I/O was definitely better with less 'nodes on the host. Does anyone with a bigger Linode have time to do the test?

_________________
/ Peter

I think you will find that /dev/urandom is the speed limiting factor in this test. Quite a few CPU cycles go into reading from that device.

Indeed, /dev/zero is probably what you want to use for disk benchmarks.

Stever - you're correct - I just checked the script for 'naughties' before I ran it, didn't give much thought to what it was using as it's input.

Guzpaz - yep - using /dev/zero pumps the disk I/O rate up to 30.6 MB/s on a 360.

_________________
/ Peter

This command may be of interest to benchmark your disk speed:

hdparm -tT /dev/xvda