What you've missed quoting from my FAQ is:
===
This requires 8 Mega (1024*1024) inodes if we want to store all
possible device nodes.
===
Note the use of the word "possible". This was not meant to imply "in
use".
> Max number of Hosts -- 64
> Max number of Channels -- 16
> Max number of Ids -- 16
> Max number of lun's -- 8
> Max number of partitions -- 64
>
> So a totally maximum system would have:
> 64 hosts * 16 channels per host == 1024 SCSI channels
[...]
> The machine would have a total of 122880 disks.
I guess I should make the FAQ clearer. I'm not talking about maximally
configured systems with 1024 SCSI channels and roomfuls of discs. What
I mean by 8 mega possible inodes is that this is the number of inodes
you would theoretically need such that you could mount any of your
discs without needing to create new inodes.
As an example, consider a (large) system with 64 channels. Since you
(the OS distributor) don't know if they will be spread across 64 hosts
each with a single channel, or 4 hosts each with 16 channels, you
*have* to create inodes for all possible combinations, which is 1024
channels. You use the same argument for the other components (IDs,
LUNs and partitions).
The essential point in the figure of 8 mega inodes is that having a
/dev that big is impossible, so you *have* to have some kind of
automagic device node generation, be it devfs, scsidev or some
such. The alternative is to not use automagic device node generation,
but that can become really painful on a large system with many
channels if you have to move discs around. For example one of your
controllers has died, so you quickly redistribute the discs across the
others and then wait for a replacement part to come in. You *really
don't* want to create those inodes manually (I consider MAKEDEVing
a manual operation).
[...]
> Therefore, based on the above facts, the maximum SCSI configuration
> that you keep referring to is a myth.
>
> Since a maximum SCSI configuration based on Richard Gooch's own numbers
> from the dev_fs FAQ is in fact a mythical machine, I would ask that
> you begin to deal in facts as you demand of us and particularly me.
I hope my explanation above clarifies things. I am certainly *not*
considering systems which are "maximally configured" in the way you
described. I am talking about the maximum number of *possible*
configurations.
Regards,
Richard....
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html