[Info-vax] Looking for suggestions for new $GETDVI item codes

[Bob Gezelter]

On Thursday, January 15, 2015 at 9:35:57 PM UTC-5, JF Mezei wrote:
> On 15-01-15 20:29, Simon Clubley wrote:
> 
> > For example, the filesystem needs to know the location of files by
> > LBA. You can work around that by using blocklets,
> 
> How do other opearting systems handle devices with different physical
> block sizes ? Do they use virtual block addresses ?
> 
> Then again, if you synthetize 512byte block numbers, you still need to
> increase the storage to 64 bits so that at run time, the system can
> fully address each virtual block in a very large file.
> 
> For the long term, one would want a system that is independent of
> physical block size so that as devices evolve, you don't have to fully
> rewrite the OS every time.
> 
> In the case of RMS indexed files where they use relative file address
> for pointers in the index, is that based on assumption of 512byte blocks
> ? (I forgot how the index pointers to bucktets work, but recall buckets
> can be more than 512 bytes)
> 
> Obviously, any change at the driver level to support new devices would
> have to be transparent to existing RMS indexed files that are just
> copied from old system to the new one. (or release notes must specify
> that a convert must be done for every indexed file)

JF,

Were it to be that simple. One of the strengths of OpenVMS is that file formats eschew hardware-dependent addressing (e.g., cylinder/track/sector) in favor of the universal 512 bytes/block convention. For examples of how hardware dependent addressing can cause pain, see the history of IBM's OS/360, which used TTR (track/record addressing) on devices with different track sizes. Every device change lead to months of work restructuring files.

Finding a good answer is difficult. The problem with the increased sector sizes is that the update quantum of 512 bytes is an underlying principal. It can be faked (e.g., with buffering and caching), but it does create opportunities for a wide range of problems.

The simplest case is that of sequential files, where it may be possible to finesse the issue using the FILES-11 cluster size and restrictions on RMS buffer sizes.

- Bob Gezelter, http://www.rlgsc.com