on transperancy

[Arun kumar]

can anyone give me some points justifying "Transperancy is the key
factor in desing issue".thanks for reading & pls give me some points

[Crosbie Fitch]

The more transparent a distributed system is, the easier it is to use.

??

Without transparency it's a pain in the arse.

Transparency is the holy grail.

You might as well ask to justify "Ergonomics is the key issue in user
interface design"?

Post by Arun kumar
can anyone give me some points justifying "Transperancy is the key
factor in desing issue".thanks for reading & pls give me some points

[Ken Hagan]

Post by Crosbie Fitch
Transparency is the holy grail.

Unobtainable?

Post by Crosbie Fitch
You might as well ask to justify "Ergonomics is the key issue in user
interface design"?

Indeed, but I suspect the OP doesn't have any control over the topic. :)

To the OP: rather than trying to justify "transparency at all costs",
perhaps you'd consider the proposition that the cost of transparency
is a key design issue. As Crosbie points out, it is *obvious* that
transparency is nice to have. It only becomes a key design issue if
it is difficult to provide and, in consequence, the designers spend
most of their time trying to figure out an architecture that provides
it.

[russell kym horsell]

Post by Ken Hagan

Post by Crosbie Fitch
Transparency is the holy grail.

Unobtainable?

[...]

I guess that depends on the context of "transparancy to whom (sic:)".
If you want something that's do-able, location transparancy for
files and processes to portable application programs is pretty easy to handle.
A simple user-level RPC driven transparent filesystem impl can even run within
(say) 10% of "normal" speeds. Obviously -- don't turn every "open" into an RPC.
Go with the 'NIX access patterns, etc.

While I'm all for transparency for development work and personal
hacking at home (some people might like to view the relevant pix of "baby"),
it turns out such things can be a hindrance.

A little project I was running for a couple of years eventually had
to back down from a home-rolled distributed 'nix with "application level
transparency" (i.e. enough to run cvs, httpd, emacs and most of the usual
app tools, as well as the "dumber" user-origined number crunching stuff)
back onto "standard" Linux + MPI because the killer app in question wasn't
dumb enough to use the system's transparent placement of files and processes,
and yet was too dumb to understand the host algorithms were trying to
be helpful. Net result -- until back-off --

think of 2 meshing gears with in-congruent numbers of teeth.

[Edward A. Feustel]

Post by Arun kumar
can anyone give me some points justifying "Transperancy is the key
factor in desing issue".thanks for reading & pls give me some points

Transparency is only ONE of a number of issues that are critical in
distributed applications.
Which issue is most important depends on your perspective as to what MUST be
achieved.

Another question is: transparency to whom?

The RM-ODP says that the distributed application needs it, but the engineers
who build
it (and want to optimize its performance) must deal with the issue of
location.

Regards,
Ed

[cr88192]

Post by Edward A. Feustel

Post by Arun kumar
can anyone give me some points justifying "Transperancy is the key
factor in desing issue".thanks for reading & pls give me some points

Transparency is only ONE of a number of issues that are critical in
distributed applications.
Which issue is most important depends on your perspective as to what MUST be
achieved.

(hmm, I never came up with this, maybe this is why my stuff never goes
anywhere...).

Post by Edward A. Feustel
Another question is: transparency to whom?
The RM-ODP says that the distributed application needs it, but the engineers
who build
it (and want to optimize its performance) must deal with the issue of
location.

in the case of a concurrent language, I think many of the issues should be
left with the language designer...


but a more general question imo is: what is the point of distributed
computing? this is less obvious.
to many would be that it is the evaluation of large and centralized
computations. this seems horridly unimaginative, but, somehow, is the only
really signifigant use for it at present...

similar is: what exactly is distributed computing? similarly, less obvious.
one can ask about more specific subjects, ie, concurrent systems, parallel
systems, "web services", p2p, ... but this still does not answer anything.
maybe in this respect "distributed computing" is in concept similar to
"artificial intelligence", namely in that it only really retains this title
until a specific form or use is found, then it gets renamed.