Xref: utzoo comp.dcom.lans:2251 comp.protocols.tcp-ip:5967
Path: utzoo!attcan!uunet!lll-winken!ames!haven!mimsy!tank!shamash!com50!mscunx!moria!tdh
From: tdh@moriaMoria.Sp.Unisys.Com (Thomas Hintz)
Newsgroups: comp.dcom.lans,comp.protocols.tcp-ip
Subject: Re: basic info about Network File System (NFS)
Summary: NFS can be implementation specific
Keywords: NFS record level I/O char codes mainframes
Message-ID: <341@moriaMoria.Sp.Unisys.Com>
Date: 11 Jan 89 17:13:11 GMT
References: <369@siswat.UUCP>
Organization: Unisys: CSI Engineering, Eagan MN
Lines: 26

In article <369@siswat.UUCP>, buck@siswat.UUCP (A. Lester Buck) writes:
> Can NFS handle the idea of records in files, as on mainframes?
> ...
> Is this done by putting some record layering code in the
> file pipeline, or what?  References would be great.
> Can NFS handle different character codes (i.e., ASCII to/from
> EBCDIC) between machines?  Not the simple case of all text
> or all binary, but say a database with binary fields and
> A. Lester Buck		...!uhnix1!moray!siswat!buck

NFS doesn't have 'hooks' persay.  NFS files appear as normal data
files to the system.  A NFS implementaion can choose to filter files
in special ways depending on some context of the file (ie. the same file
named several ways could trigger different filters).  I reality, I
don't know an NFS that does that, but then I don't know IBM's NFS.

I may be porting NFS to UNISYS 1100 series mainframes soon.  The same
conversion problems exist: 1100 data files occur in many forms.  If we
do this, we may choose to allow filtering of files by both name context,
and by data content since 1100 files also have 'SDF' words to describe
each record.  Sounds messy.
-- 
- - - -
Thomas D. Hintz				(612) 687-2684
UNISYS					tdh@moria.sp.unisys.com
Network Engineering			..!com50!mscunx!moria!tdh