Benchmarking: IOZone (CERN)

Prerequisites

XFS & lsscsi need installing:

yum install xfsprogs.x86_64 lsscsi

As does IOZONE:

yum install iozone

Preparing the data filesystems

All available block devices in the attached disk enclosure will be used for this test. None of the drives in the front-end server will be used.

Get the list of available block devices

Use lsscsi to identify the drives to be tested (in this case, /dev/sdc to /dev/sdr):

[root@nodea ~]# lsscsi
[0:0:0:0]    enclosu Promise  J630s            060=  -       
[0:0:1:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdc 
[0:0:2:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdd 
[0:0:3:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sde 
[0:0:4:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdf 
[0:0:5:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdg 
[0:0:6:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdh 
[0:0:7:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdi 
[0:0:8:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdj 
[0:0:9:0]    disk    ATA      ST6000NM0024-1HT SN02  /dev/sdk 
[0:0:10:0]   disk    ATA      ST6000NM0024-1HT SN02  /dev/sdl 
[0:0:11:0]   disk    ATA      ST6000NM0024-1HT SN02  /dev/sdm 
[0:0:12:0]   disk    ATA      ST6000NM0024-1HT SN02  /dev/sdn 
[0:0:13:0]   disk    ATA      ST6000NM0024-1HT SN02  /dev/sdo 
[0:0:14:0]   disk    ATA      ST6000NM0024-1HT SN02  /dev/sdp 
[0:0:15:0]   disk    ATA      ST6000NM0024-1HT SN02  /dev/sdq 
[0:0:16:0]   disk    ATA      ST6000NM0024-1HT SN02  /dev/sdr 
[1:0:0:0]    disk    ATA      Hitachi HUA72302 MK7O  /dev/sda 
[2:0:0:0]    disk    ATA      Hitachi HUA72302 MK7O  /dev/sdb

Create an XFS filesystem on each drive

The basic command for this is:

mkfs.xfs -l version=2 -i size=1024 -n size=16384 -d su=64k,sw=1 <block device>

So can be scripted as follows:

for i in {c..r};
do
	mkfs.xfs -l version=2 -i size=1024 -n size=16384 -d su=64k,sw=1 /dev/sd${i} &
done

Create mount points

Create mount points in /srv/castor/XY where XY ranges from 1 to the number of available block devices. For example, it will be 16 on a 16-bay disk enclosure.
This can be scripted as follows:

mkdir /srv/castor
for i in {1..16};
do
	mkdir /srv/castor/${i}
done

Mount filesystems to mount points

Mount these filesystems using the following options: logbufs=8,logbsize=256k,noatime,swallow,inode64
This can be scripted to add mount point to /etc/fstab as follows:

n=1

for i in {c..r}
do
	echo "/dev/sd${i}	/srv/castor/${n}	xfs logbufs=8,logbsize=256k,noatime,swalloc,inode64" >> /etc/fstab
	((n++))
done