Re: [ADSM-L] File serving with TSM for Space Management

2012-04-10 05:09:28
Subject: Re: [ADSM-L] File serving with TSM for Space Management
From: Richard Sims <rbs AT BU DOT EDU>
Date: Tue, 10 Apr 2012 05:02:41 -0400
On Apr 9, 2012, at 2:22 PM, Paul Zarnowski wrote:

> Does anyone who is using TSM for Space Management (aka HSM) know if it can be 
> used to share out filesystems using NFS (or CIFS)?  Or are there timeout 
> problems that make this unworkable?

You can, and we do it via NFS - but it can be a nightmare at times. Both HSM 
and NFS operate as kernel extensions: any problems thus become very severe. We 
had someone who thought it a good idea to have an HSM file system NFS-mounted 
to an FTP server for data feeding. That resulted in a hung FTP server system 
whenever data was being written faster than it could be migrated on the HSM 
server. Over the past weekend we had a situation where a user thought it 
reasonable to copy a movie file larger than the 64 GB HSM file system into that 
area: the file system was wedged, and because it was NFS-served as well, NFSd 
was hung; and the incoming files could not migrate because they were in an open 
state, even after the writing process was killed off: we had to reboot the 
server (which in turn somehow incited the failure of a IBM RAID adapter card 
and a marathon recovery effort that I'm just recovering from). A little known 
reality about HSM is that the space for a file must consist of contiguous 
blocks: you can have an HSM file system that is like 85% full as incited by a 
large file being written, and writing can proceed no further, because the file 
system is fragmented. Because of this, for file systems which get large files 
from users, we have an early morning job which forcibly migrates everything out 
of the file system to maximize free space.

But don't let me discourage you from using HSM...  :-)

     Richard Sims   enduring at Boston University

<Prev in Thread] Current Thread [Next in Thread>