vrs-development
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Vrs-development] Who's doing the distributed filesystem?


From: Bill Lance
Subject: Re: [Vrs-development] Who's doing the distributed filesystem?
Date: Mon, 29 Apr 2002 05:52:58 -0700 (PDT)

--- Eric Altendorf <address@hidden> wrote:
> 
> The more I think about it, the trickier our
> theoretical distributed 
> filesystem sounds.
> 
> It seems to me (though I could easily be wrong) that
> we're basically looking 
> for a dynamically scalable, highly distributed,
> low-bandwidth, secure and 
> encrypted, interleaving and redundancy
> -parameterizable RAID system. 

That's my thought here also, Eric.  It's basically a
type 10 RAID system with encrypted blocks and spread
accross the network

 We've 
> got to think about semantics... transactional
> guarantees... performance...
> 
> My earlier analogy to the Andrew file system was way
> off.  AFS replicates 
> whole files, and has session semantics (updates to
> files are not visible 
> until the file is closed).  What we're talking about
> here is many times more 
> complex.  I'm not sure if it's practical...  we
> might be able to do it but I 
> don't know if we could get decent performance.
>

We are indeed assuming a great deal of overhead here. 
But several points should be made.  One, 'decent
performance' is always a matter of (expectations <->
resources).  As our resources spiral upward, so does
our expectations.  So we are never satisfied  :)
The other point is that our available resources DOES
continue to spiral upward.  So, if we have worked out
the logic of the problem and we don't yet have enough
resources yet to push that logic fast enough to meet
some basic threashold, we just have to wait a bit till
the resources meet our need.

Right now, we are working out the logic of it.


 
> Do we have any distributed filesystem experts here? 
> I'm sitting here 
> re-reading the filesystems chapters from
> Silberschatz and Galvin and next 
> I'll be working through Gray and Reuter's
> transaction processing book, but 
> that's not going to make me an expert. :-)
> 

I don't think we do have such an expert aboard yet, so
go for it!  We have been working from some basic
principles and wish-lists so far.

If this is the area that you would like to focus on,
that would be great.  We are taking a top-down design
approach.  Where we are at so far is in the Docs in
the Repositiry Manager section of both the 'Local Data
Server Design Detail' page and the Transactions page.

The Design Detail part is an architectural overview. 
Does any part of that  need further clarification or
need changes?  The 'Transactions' section details the
process interaction of the LDS elements.  That needs
much more thought.

As soon as Chris has the GW stuff working, we can then
start writing working code.  At this point, we are
planning on the initial logic to be in perl for
prototyping.  But the nature of GW gives us a lot of
flexibility in what modules are written in what.


 
> PS: If I'm being annoying or asking questions that
> have already been 
> addressed in archives or something, please tell me. 
> I want to be a help, not 
> a distracting loudmouth. :-)
> 

Absolutly no problem, Eric.  There can be no answers
without questions.  And I am most certain that we have
only begun to ask the right questions.


__________________________________________________
Do You Yahoo!?
Yahoo! Health - your guide to health and wellness
http://health.yahoo.com



reply via email to

[Prev in Thread] Current Thread [Next in Thread]