How to configure diskless disk setup when using Torque

    

Torque behaves strangley when tmpfs is specified as "0" in the disksetup. The setting "0" means "unlimited" - eg, use as much of the memory as is necessary to contain the image. But Torque sees this as a full file system.

There is an easy workaround.


 

To eliminate this problem, modify the disksetup to use something other than zero. We recommend that you use 90%. In this case, the maximum amount of memory that tmpfs can use is 90% of the total installed. In practice, it will only use what it needs to contain the image, and the rest will be available for user processes.

Here's how to do it using the Bright cmsh in 5 fast and easy steps.

First...

[root@atom-head1 ~]# cmsh
[atom-head1]% category use default
[atom-head1->category[default]]% get disksetup

<?xml version="1.0" encoding="UTF-8"?>
<diskSetup xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
   <diskless maxMemSize="0"/>
   <device>
     <blockdev>/dev/sda</blockdev>
     <partition id="a1">
       <size>max</size>
       <type>linux</type>
       <filesystem>ext3</filesystem>
       <mountPoint>/scratch</mountPoint> <mountOptions>defaults,noatime,nodiratime</mountOptions>
     </partition>
   </device>
</diskSetup>

Now change it to use up to 90% as follows:

[atom-head1->category[default]]% set disksetup

<?xml version="1.0" encoding="UTF-8"?>
<diskSetup xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
   <diskless maxMemSize="90%"/>
   <device>
     <blockdev>/dev/sda</blockdev>
     <partition id="a1">
       <size>max</size>
       <type>linux</type>
       <filesystem>ext3</filesystem>
       <mountPoint>/scratch</mountPoint> <mountOptions>defaults,noatime,nodiratime</mountOptions>
     </partition>
   </device>
</diskSetup>

Now commit your changes so that they take effect.

[atom-head1->category*[default*]]% commit


Please clear the node-level disk setup on n1.

Once cleared it will use the category-level disk setup.

[atom-head1->category[default]]% device use atom01
[atom-head1->device[atom01]]% clear disksetup
[atom-head1->device*[atom01*]]% commit

Finally, reboot your compute nodes to change the partitions.

[atom-head1->device[atom01]]% exit
[atom-head1->device]% reboot -c default

 

You're done.

 

High Performance Computing eBook