[Info-vax] what is a good value for /CLUSTER_SIZE? (INITIALIZE and qualifiers)

[Hein RMS van den Heuvel]

On Feb 5, 1:02 pm, hel... at astro.multiCLOTHESvax.de (Phillip Helbig---
undress to reply) wrote:

> Up until now, I have usually taken the defaults of INITIALIZE, except
> where something else (e.g. /SYSTEM) was absolutely necessary (and, of
> course

The defaults used to be 'odd'. Literally.
They've become better: Even, multiples of 16.
The driving factor for the is that many storage controllers and the
XFC like that.
And RMS (sequential file) buffers defaults to 16 (old) or 32 (new)
blocks.
All of that playing together you stack the deck just a little to
improve the odds that things, like fragments, line up nicely.
No more, no less.

I specify /HEADERS as much larger than the default; I use 50,000
> per GB as a rule of thumb).

So you expect an average file size of about 40 blocks,
and your are fine pre-allocating 450,000 headers (1/4 GB) on a 9GB
drive. Fine by me.

>  Since I plan to do some stuff with DDS and DVE soon,

DVE = Dynamice Volume Expansion. DDS = ?

> I need  to specify /LIMIT.  This will give me a cluster size of 8 unless I
> specify something else.

So? 8 is fine, but if you don't like it specify your own.
Personally I like larger and powers of 2 : 32? 128?

The price of large clusters is wasted space in the last cluster of a
file.
You may guestimate that on average to be 1/2 a cluster per file
For small files, I find the waste is closer to the clustersize minus 3
or so.
So for those 450,000 files, how much are you willing to 'waste' in
roundup?
With a cluster size of 16 that could be 2GB with a cluster size of 8
it would be 1 GB-ish
But if the volume will only have say 10,000 files, then a clustersize
of 128 wastes 1/2 GB at the most.

The price for small clusters is potentially extra (free space)
fragmentation and just more 'work' to manage everywhere.

Pick your poison. Don't pick 42 not 17. Pick 32 or 16.

> Should I stick with the default implied by /LIMIT, i.e. a cluster size

NEVER accept the default if you worry about it or try to understand
it.
Always specify, if only for documentation

> the default of 8 used by /LIMIT is usually significantly smaller than what
> the default cluster size otherwise would be.  Is there a reason for
> this?  More generally, why introduce a new default with /LIMIT at all?

The real answer? Oracle!  To support their build environment.

To truly exploit the new larger BITMAP.SYS opportunity making large
enough to expand the volume and not be 'stuck' with the old default of
just 1 million clusters. (which btw implies an absolute maximum of 1
million files... given at least one cluster per file)

Also, this /LIMIT came to be when HARD_LINKS were designed.and those
are implemented as silly, dinky, 1 block, files.
Those create maximum roundup wastage of cluster-size minus 1 for each
link.

Know your application: files count, file sizes and pick you poison, or
just don't sweat the small stuff!

Cheers,
Hein