[Info-vax] VMS - The new file system. What do we know about it?...

[Dirk Munk]

Michael Moroney wrote:
> Dirk Munk <munk at home.nl> writes:
>
>> Modern disks of course use the Advanced Format structure, with 4kB
>> blocks/sectors. Using arbitrary 512 byte blocks is possible, but not
>> very good for performance. Is that taken into consideration?
>
> That will have to be a separate project.  From what someone here told me
> other OS's (Linux, Windoze) still use 4K block drives in 512 byte block
> emulation mode.  Can anyone verify/refute this?
>
> Also, most VMS IO on itanics is through SAN storage servers, and they
> are the ones that have to directly deal with 4K blocks.  I don't know
> if storage servers themselves present themselves as using 4K blocks.
>
> Of course there are directly attached SCSI drives that could benefit,
> plus x86 is a different beast.
>
Linus and Unix use logical block sizes of 4kB, or multiples of 4kB. 
Windows uses cluster sizes of 4kB to 64kB. As far as I'm aware the 
minimum IO size of these operating systems is one logical block / cluster.

It is very important that sector borders correspond to logical block / 
cluster borders. If they do not correspond, then you get performance 
problems. Suppose you write one 4kB logical block that starts in the 
middle of a 4kB sector, then the disk has to read two sectors, divide 
the logical block over the two sectors, and rewrite both sectors.

 From this we can conclude that the problem has the be solved by 
partitioning the right way. The old MBR partition used a partition table 
of 63 512byte blocks, so if you didn't do anything the first partition 
(and most likely other partitions as well) would start in the middle of 
a 4kB sector.

I made sure the first partition always started at block 2048 (1MB), and 
that every partition started at a MB border. Of course using the 
Cylinder/Head/Sector method for partition is completely ludicrous, never 
use it.