*BSD News Article 80130


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!news.wildstar.net!imci5!pull-feed.internetmci.com!newsfeed.internetmci.com!feed1.news.erols.com!howland.erols.net!www.nntp.primenet.com!nntp.primenet.com!news.mathworks.com!fu-berlin.de!irz401!orion.sax.de!uriah.heep!news
From: j@uriah.heep.sax.de (J Wunsch)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: SCSI tape block size
Date: 6 Oct 1996 18:09:29 GMT
Organization: Private BSD site, Dresden
Lines: 33
Message-ID: <538skp$90e@uriah.heep.sax.de>
References: <8720ffxq0j.fsf@plm.xs4all.nl> <536grc$ba@uriah.heep.sax.de>
  <537u8o$b8@mscu.snafu.de>
Reply-To: joerg_wunsch@uriah.heep.sax.de (Joerg Wunsch)
NNTP-Posting-Host: localhost.heep.sax.de
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 8bit
X-Newsreader: knews 0.9.6
X-Phone: +49-351-2012 669
X-PGP-Fingerprint: DC 47 E6 E4 FF A6 E9 8F  93 21 E0 7D F9 12 D6 4E

matthias@mscu.snafu.de (Matthias Schuendehuette) wrote:

> > (Note that 64 KB is the largest blocksize currently supported inside
> > the kernel by physio(), so there's no use in making the parameter
> > above larger.)
>  
> There was a rather large thread in de.comp.os.unix on reading
> tar-archives from SGI machines which have default blocksizes of 256 KB.

I've read it (and posted there -- as you probably know).  And yes,
FreeBSD currently cannot read these tapes, that's where we've got
aware of the problem at all.

> Not to acknowledge this nonstandard behaviour of IRIX but would it be a
> great effort to extend the max. blocksize in physio() and what would be
> the drawbacks of doing so?

The limitation arises out of the simple fact that quite some SCSI host
adapters do not support more than 16 scatter/gather segments.  In an
VM architecture, you need one scatter/gather segment per logical page
in the worst case.  This makes 16 * 4 KB = 64 KB.  Relinguishing this
limit would require a partial rewrite of some of the lower-level
drivers (they probably have to use bounce buffers or something like
that).

Despite of this, i still think SGI is plain insane in this...

-- 
cheers, J"org

joerg_wunsch@uriah.heep.sax.de -- http://www.sax.de/~joerg/ -- NIC: JW11-RIPE
Never trust an operating system you don't have sources for. ;-)