> > The same line of reasoning applies to a number of other areas.  AFS
> > solves no problems that cannot be solved other ways, using local disks,
> > NFS, etc., albeit requiring more human intervention.
> 
> I want to see the exact same terabyte file system on 20,000 
> machines, all of which are actively using it at the same time.
> 
> How do I do this with NFS?

I want the same thing.  Well, not quite 20,000 systems - I want
to have access to the same data from each of my company's sites
anywhere in the world.  But, the applications are well-known enough
that special case solutions suffice.


Here's how it seems to be evolving:

a) some significant fraction of that terabyte that people want to
        share is version controlled "source code", where source code
        means C/C++, but also HTML, documents, ASCII text, etc.

        Let's say, 200GB.

        What people do is throw up a CVS server on their local systems
        - 200GB of disk.  When the server gets overloaded, they buy a bigger
        one. But, the decision of when to buy is left to the departments,
        not central IT.

        Now, probably 5X that space - 1 TB - is used for local images
        of the source code.


b) Much of the remaining 800GB is read-mainly, changing infrequently,
        used by CAD tools and the like being run over a network queuing
        program.  Reference inputs, traces, etc.

        Let's say, another 700GB.

        At each of your corporate sites, you put up a big NFS server.
        You replicate - via scripts that FTP, rsync, or the like - the
        read-mainly data to each of these sites. Again, maybe a 10X increase
        in total disk space.

c) Finally, 100GB of truly read-write shared data.
        This is where AFS shines.


Obviously bogus numbers, but you get the picture: ad-hoc replication.
Increasing disk space requirements from 1TB of AFS space to 9TB of
non-AFS space. But, a lot of that non-AFS space is with PC class
rather than server class disks.  Raw disk cost for PC class disks 
is 4K$/TB. 9TB costs, raw disk cost, maybe 40,000$US. That's less
than one AFS administrator.

What about cost of ownership?  Sysadmins for all those extra systems
(well, you probably have them anyway). Backup operators?  The dirty little
secret of more and more sites is that the vast majority of disks are not 
backed up.


I think this is utterly and totally bogus.  I don't think this is good.
But, I am trying to understand the motivations that have led to this 
sort of situation - e.g. where I return to a group I left 4 years ago,
where the AFS space has increased not at all, but where the local disk 
space on PCs has increased 100-fold - and I can begin to see why this
is happening.


Reply via email to