biolocal:Nodedisk - Revision history

Mauro at 11:05, 28 October 2009

Mauro — Wed, 28 Oct 2009 11:05:12 GMT

←Older revision		Revision as of 11:05, 28 October 2009
Line 26:		Line 26:
	,104127,L		,104127,L
	EOF		EOF
-	/sbin/mkswap /dev/hda1; /sbin/mkswap ~~/dev/hda2; /sbin/swapon /dev/hda1; /sbin/swapon~~ /dev/hda2;	+	/sbin/mkswap /dev/hda1; /sbin/mkswap /dev/hda2;
	/sbin/mkfs -t ext3 -L scratch /dev/hda3; /bin/mount LABEL=scratch		/sbin/mkfs -t ext3 -L scratch /dev/hda3; /bin/mount LABEL=scratch
	fi		fi
	fi		fi
		+	/sbin/swapon /dev/hda1; /sbin/swapon /dev/hda2;
	</PRE>		</PRE>

Gianluca at 12:54, 4 August 2009

Gianluca — Tue, 04 Aug 2009 12:54:22 GMT

←Older revision		Revision as of 12:54, 4 August 2009
Line 7:		Line 7:
	The new CentOS 5 cluster installation is based on labels for disk partitioning.		The new CentOS 5 cluster installation is based on labels for disk partitioning.

-	A script is installed in /usr/local/sbin/build_disk.sh to create at boot time the disk partitioning if the expected partitions ~~doesn~~'t exist.	+	A script is installed in /usr/local/sbin/build_disk.sh to create at boot time the disk partitioning if the expected partitions don't exist.

	<PRE>		<PRE>
Line 32:		Line 32:
	</PRE>		</PRE>

-	This script is called by rc.local and works only if the partition label "scratch" ~~doesn0~~'t ~~exist~~ and any hda partition is already mounted.	+	This script is called by rc.local and works only if the partition label "scratch" isn't found and any hda partition is already mounted.

	sfdisk takes some command from the std input to create the partitions:		sfdisk takes some command from the std input to create the partitions:
-	from 0 cylinder, with size 4161 cylinders, of type swap	+	*from 0 cylinder, with size 4161 cylinders, of type swap (0,4161,S)
-	from last cyl, with size 4161 cylinders, of type swap	+	*from last cyl, with size 4161 cylinders, of type swap (,4161,S)
-	from last cyl, with size 104127 cylinders, of type linux	+	*from last cyl, with size 104127 cylinders, of type linux (,104127,L)

	Finally the partitions are created and mounted.		Finally the partitions are created and mounted.

	''Note : The swap partition should always exist, otherwise the scheduler wouldn't get the node (it looks for nodes with free swap space)''		''Note : The swap partition should always exist, otherwise the scheduler wouldn't get the node (it looks for nodes with free swap space)''

Gianluca: New page: Each node should be initialized with the following partitions on the local disk: * swap1 - 2Gb * swap2 - 2Gb * scratch - 50Gb The new CentOS 5 cluster installation is based on labels for...

Gianluca — Tue, 04 Aug 2009 12:51:15 GMT

New page: Each node should be initialized with the following partitions on the local disk: * swap1 - 2Gb * swap2 - 2Gb * scratch - 50Gb The new CentOS 5 cluster installation is based on labels for...

New page

Each node should be initialized with the following partitions on the local disk:
* swap1 - 2Gb
* swap2 - 2Gb
* scratch - 50Gb

The new CentOS 5 cluster installation is based on labels for disk partitioning.

A script is installed in /usr/local/sbin/build_disk.sh to create at boot time the disk partitioning if the expected partitions doesn't exist.

<PRE>
cat /usr/local/sbin/build_disk.sh

#!/bin/sh

/sbin/findfs LABEL=scratch 2>1 > /dev/null;

if [ $? -ne 0 ];
then
/bin/mount | grep hda
if [ $? -ne 0 ];
then
/sbin/sfdisk /dev/hda << EOF
0,4161,S
,4161,S
,104127,L
EOF
/sbin/mkswap /dev/hda1; /sbin/mkswap /dev/hda2; /sbin/swapon /dev/hda1; /sbin/swapon /dev/hda2;
/sbin/mkfs -t ext3 -L scratch /dev/hda3; /bin/mount LABEL=scratch
fi
fi
</PRE>

This script is called by rc.local and works only if the partition label "scratch" doesn0't exist and any hda partition is already mounted.

sfdisk takes some command from the std input to create the partitions:
from 0 cylinder, with size 4161 cylinders, of type swap
from last cyl, with size 4161 cylinders, of type swap
from last cyl, with size 104127 cylinders, of type linux

Finally the partitions are created and mounted.

''Note : The swap partition should always exist, otherwise the scheduler wouldn't get the node (it looks for nodes with free swap space)''