is there a lisp app that can do clustering and make 100 pc appear as
1 big computer all done in lisp over tcp/ip in userspace? all
resources go into a cloud?
On Thu, 13 Nov 2008 11:37:12 -0800 (PST), gavino <·········@gmail.com>
wrote:
>is there a lisp app that can do clustering and make 100 pc appear as
>1 big computer all done in lisp over tcp/ip in userspace? all
>resources go into a cloud?
You can use shared virtual memory in any language.
I/O resources and remote code execution are another matter. Cloud
computing is mainly hype at this point ... only a handful of the
"network operating systems" I've heard of are really practical to use,
but all of them are "experimental" and none are designed to work long
distance over the internet.
George
GN> I/O resources and remote code execution are another matter. Cloud
GN> computing is mainly hype at this point ...
it depends on what you call cloud computing.. typically people mean
things like Amazon Web Services, Google Apps etc, and this things
are already quite practical and useful.
i'd say what cloud computing definitely is not is "abstracting away
network" -- it is more like making global network a vital part of
infrastructure.
as i understand it, the old approach was to use either a "big iron"
or clustered solution -- expensive, high-end and tightly coupled hardware.
this allows to make scalability transparent for applications. at the cost
of high expenses, of course.
in contrast new approach of "cloud computing" is to make use of
networking explicit (then it becomes less sensitive to latencies etc.
because you see it), use lots of cheapo boxes etc. this allows to
drastically cut expenses to scale to high levels, but solution becomes
less reliable and more complex on application level.
On Fri, 14 Nov 2008 16:09:00 +0200, "Alex Mizrahi"
<········@users.sourceforge.net> wrote:
> GN> I/O resources and remote code execution are another matter. Cloud
> GN> computing is mainly hype at this point ...
>
>it depends on what you call cloud computing.. typically people mean
>things like Amazon Web Services, Google Apps etc, and this things
>are already quite practical and useful.
>
>i'd say what cloud computing definitely is not is "abstracting away
>network" -- it is more like making global network a vital part of
>infrastructure.
Yes. The problem is that the web services model does not easily
expand beyond the traditional reach of the database cluster. It works
well for reactive applications but not as well for heavy computing,
and the browser model is not applicable for many types of
applications.
>as i understand it, the old approach was to use either a "big iron"
>or clustered solution -- expensive, high-end and tightly coupled hardware.
>this allows to make scalability transparent for applications. at the cost
>of high expenses, of course.
I'm not talking specifically about cluster OSes - they serve their
purpose, but as you indicated, are frequently limited to related
hardware.
The idea of the distributed OS was that "the network is the computer".
You don't log into a host but into a system and you don't need to be
particularly aware of where your applications actually run. From the
perspective of the user, everything works the same (or as nearly as
possible) regardless of the user's location.
Probably the best attempt was Amoeba. Amoeba supported dissimilar
architectures, could decide (based on program metadata) which
architecture were supported and which were preferable for an
application (or slice of one) and could migrate apps for load
balancing or to be near needed I/O devices or to boost performance on
more suitable processors. All devices (including processors) were
uniquely named and I/O could be seamlessly routed between any
processor and device.
>in contrast new approach of "cloud computing" is to make use of
>networking explicit (then it becomes less sensitive to latencies etc.
>because you see it), use lots of cheapo boxes etc. this allows to
>drastically cut expenses to scale to high levels, but solution becomes
>less reliable and more complex on application level.
Cloud computing makes a lot of the same claims as distributed OSes,
but it puts the burden of location transparency on the application
rather than on the operating system.
George
GN> Yes. The problem is that the web services model does not easily
GN> expand beyond the traditional reach of the database cluster. It works
GN> well for reactive applications but not as well for heavy computing,
Amazon Elastic Compute Cloud (EC2) works fine for heavy computing.
web services are not really involved there -- basically on demand they give
you access to a (virtualized) box, and you do whatever you want with
it -- control it via ssh, connect to other such boxes..
GN> Cloud computing makes a lot of the same claims as distributed OSes,
GN> but it puts the burden of location transparency on the application
GN> rather than on the operating system.
i guess if it's getting popular simlistic approach of amazon web services
and
stuff like that is enough for practical purposes, while switching to real
distributed
OS like Amoeba is associated with additional complexities and overheads, and
so people are not eager to use it.
however, i think if distributed computing stuff will go mainstream,
eventually people
will get bored implementing it in ad-hoc way, and will demand distribution
being
supported on OS level. Microsoft announced that they have research projects
in this area (Midori) and, it seems, they are quite serious with it, so
eventually
we are likely to see this stuff in mainstream.