Multiple JFS2 in a devclass type file.

ypcat

ADSM.ORG Member
Joined
Oct 30, 2007
Messages
130
Reaction score
1
Points
0
Location
Toronto Canada
Hello Folks,

I have a question so I'm going to just come right out and ask. In a large TSM instance there used to be a bug. If you string filesystems together in the devclass definition and delimit them by comma it would work as a single storage pool until one of the file systems filled up. After that it would no longer create scratch volumes in the new (second) file system - based on the maxscratch parameter. You would have to create new volumes manually. Has that been fixed?

I was forced to create a string of very large pools to accommodate things based on JFS2 - my only choice currently so allocating resources, reporting, etc becomes a pain.

I was wondering if anyone here was stringing 32TB max JFS2's together in a single devclass (type=file) and if you have a full JFS2, does it automatically create scratch in the second file system? I'd like to consolidate my TSM storage pools.

Thanks in advance.
 
It's supposed to use all the filesystems in a sort of round robin fashion. This helps with performance and the usage of each filesystem grows somewhat at the same pace.
 
This is what the man page says about creating separate file systems for this so I've created a devclass with directories like "/usr/local/tsm/tsmflat1,/usr/local/tsm/tsmflat2,/usr/local/tsm/tsmflat3" but I want to know if those are considered separate file systems as per the man page recommendation below.

Each of these is comprised of 4 * 8TB LUNS totaling 32 TB per JFS2 (which is max in AIX). I believe the 32 TB JFS2 is a separate file system and it doesn't really matter where I mount it, but I want to be sure as I've had issues with stringing multiple file systems together under a single device class (type=file) before.

MAN PAGE EXCERPT
Tip: If you specify multiple directories for a device class, ensure
that the directories are associated with separate file systems.
Space trigger functions and storage pool space calculations
take into account the space remaining in each directory. If you
specify multiple directories for a device class and the
directories reside in the same file system, the server will
calculate space by adding values representing the space
remaining in each directory. These space calculations will be
inaccurate. Rather than choosing a storage pool with sufficient
space for an operation, the server might choose the wrong
storage pool and run out of space prematurely. For space
triggers, an inaccurate calculation might result in a failure
to expand the space available in a storage pool. Failure to
expand space in a storage pool is one of the conditions that
can cause a trigger to become disabled. If a trigger is
disabled because the space in a storage pool could not be
expanded, you can re-enable the trigger by issuing the
following command: update spacetrigger stg. No further changes
are required to the space trigger.
 
This is what the man page says about creating separate file systems for this so I've created a devclass with directories like "/usr/local/tsm/tsmflat1,/usr/local/tsm/tsmflat2,/usr/local/tsm/tsmflat3" but I want to know if those are considered separate file systems as per the man page recommendation below.
Are each of the fileystem mounted to one of those directory in the output of the AIX command "df"? If so, they are separate filesystems.

Note, you said you had 4 filesystems, but only 3 directories listed in the example. Was that an omission to keep the message shorter and just showing the pattern?
 
Just an update - it seems to be working so far creating volumes in both file systems, but we shall see when one of the file systems becomes full.
 
Are each of the fileystem mounted to one of those directory in the output of the AIX command "df"? If so, they are separate filesystems.

Note, you said you had 4 filesystems, but only 3 directories listed in the example. Was that an omission to keep the message shorter and just showing the pattern?
Yes it is in it's own listing under df - which makes sense because I had to create the filesystem, but you can never be too sure, and I did shorten it for brevity. Thanks!
 
Back
Top