xfs
[Top] [All Lists]

Re: xfsdump manual page inaccurate for -b option?

To: Jeremy Jackson <jerj@xxxxxxxxxxxx>
Subject: Re: xfsdump manual page inaccurate for -b option?
From: Tim Shimmin <tes@xxxxxxx>
Date: Tue, 8 Jun 2004 12:35:43 +1000
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <40C0D85F.9000909@coplanar.net>; from jerj@coplanar.net on Fri, Jun 04, 2004 at 04:15:27PM -0400
References: <40C0D85F.9000909@coplanar.net>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/1.2.5.1i
Hi Jeremy,

On Fri, Jun 04, 2004 at 04:15:27PM -0400, Jeremy Jackson wrote:
> Hi,
> 
> I'm reading xfsdump/common/drive_scsitape.c:calc_best_blk_and_rec_sz() 
> and it seems that the manual page is inaccurate regarding the -b switch.
> 
> It indicated a 1MiB default, but reading the code, it looks like if -m 
> is not used and if -b is not used, and the device is a local (not rmt) 
> tape device, then the block size defaults to the lesser of the drive's 
> maximum blocksize, or STAPE_MAX_RECSZ
> 
I reckon the code is a bit messy here.
I believe the man page is correct, however.
STAPE_MAX_RECSZ is 2MiB which was used on IRIX.
STAPE_MAX_LINUX_RECSZ is 1MiB for use on Linux.
And dc_maxblksz comes from mtinfo.maxblksz which comes from
STAPE_MAX_LINUX_RECSZ.
From memory linux st driver didn't support maxblksz as was the case
with the IRIX driver (only a MT_ST_BLKSIZE_SHIFT on the mt_dsreg).
We used 1MiB for Linux b/c Linux often Linux couldn't handle 2MiB.
In recent times it was pointed out that for Linux a smaller record
size than 1MiB would still give good performance.

> A description of tape records (vs tape blocks) would be helpful also.
> 
In xfsdump/doc/xfsdump.html at the end, I was wondering the same question :)

--Tim


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