Slashdot Mirror

If you look at the Global Grid Forum you will see that a derivative of SOAP (first OGSA, now WS-RF) is used as the underpinnings of the proposed Grid architecture.

Just because it is being pushed there and in OASIS does not mean that is is the right approach. Nor is EJB, IMO. EJB is designed to do distributed transactions against something that may or may not be a relational database. Grid stuff is very often computation against a large datastore; potentially chained stuff. It is not classic three tier J2EE.

It seems to me we need something like network pipes, where you can construct pipelines of computation and have the resource manager place parts of the pipe on the right machines. There is some work in the OGSA-DAI working group looking at this.

It's now used not only in Computational Science, but also in business. Butterfly http://www.butterfly.net/ uses Grid as the online game solution. And as I know, SUN comes out with SUN Grid Engine, and Oracle also has its Grid solution for business customers.

As we worked on both intranet and internet a few decades before, we now have the grid for enterprise use and also the so-called Global Grid http://www.ggf.org/.

Check out these links for details, there is a lot of work in this area. Comparing this to @Home type projects is the wrong way to go... each node on the grid can be a gatekeeper to a cluster, where a parallel job is run. The problems involved are more complex than what a centralized server/organization can take care of.

http://www.globus.org//
http://www.ggf.org//
http://www.globusworld.com//

Some details.. much more out there.
http://www.grids-center.org/news/clusterworld//
http://www-106.ibm.com/developerworks/grid/library /gr-design.html/
http://www.casa-sotomayor.net/gt3-tutorial//

Plus, lots of academic papers out there.. this is a pretty interesting subject.

Globus provides an implementation of OGSA/OGSI and soon WSRF, all protocols developed for grid computing. The Global Grid Forum meets regularly (at conferences) to decide on standardization of grid protocols.

Note that just because the grid protocols are being standardized does not mean anyone can have access to one "giant grid". It just means that heteregeneous sites are able to interoperate, authentication/authorization control is up to the participating sites' local policies and via virtual organizations which can be (simply) thought of as aggregates, the 'master' groups (but VOs can be much less coarse than that).

There is a lot of advanced work being done in this area, see:

http://www.globus.org/
http://www.ggf.org/

"Grid" as a concept is mostly just a buzzword. Oracle10g is a good example.

But Grid as a standard (under development by the GGF, OASIS and others) is something a lot more specific. What that emerging standards-compliant Grid offers is:

1. End-to-end encryption, based on certificates and public key
2. Virtual organizations, in which there is a closed "community" of systems and their users who can participate in a particular Grid (somewhat VPN-like)
3. Event-driven framework, instead of client/server or push/pull models -- this is a major win for some applications

The Globus toolkit is one messed up pile of confused & confusing software (yes, I'm running 2.4 and 3.0 and develop for it). But it will change a lot for the better over this upcoming year, if it keeps to schedule, to build on Web services (Tomcat, etc.) rather than re-inventing WS+Grid. Adding the points 1-3 above to WS will work a whole lot better, I think/hope, than re-creating most of WS in the Grid services (GS) Globus toolkit.

Bottom line: The vision/plan of standards-compliant Grid computing does offer some real advantages and promise. It's not for every application, every user and every organization, any more than, say, cluster or distributed computing is. But today's Globus and other Grid standards are in pretty early stages, and only barely useful for real tasks (i.e., see how the Teragrid fell victim to fairly mundane attack).

I think that's a really dodgy view of Grid computing. Grid computing is essentially resource/service sharing across heterogeneous nodes (i.e. different types of machines - macs/pcs/microscopes/etc). To do that, the Global Grid Forum are developing a load of standard protocols and methods for getting everything to inter-communicate.

As far as I'm aware, there is currently no standard way of purchasing CPU cycles or similar, although there are a number of working groups whose remit probably covers this.

The beauty of the Grid is more in being able to seamlessly connect to pretty much any hardware resource you want - I suspect that in reality, the actual economics will be dictated more by existing commercial agreements more than anything else.

The TeraGrid is the NSF flagship for grid computing - be it good or bad.

The Grid.org people are some of the former SETI@home people gone more general purpose.

And of course, there is The Global Grid Forum which is meeting in Chicago in a week or so. GGF is the standards behind the Globus enabled grid.

We could ask why CERN/etal couldn't have come up with a slightly more imaginary name?

We can also ask why NSF are such suckers for the last 20 years of hype from the people who have run the national supercomputer centers in the USA? Ditto congress. But that is a (sad) story for a different day.

And finally we can ask what Top500.org is going to do when people begin reporting HPL benchmarks using these things? That HPL became the standard that people are designing supercomputers around argues just how totally screwed up high performance computing really is at the moment.

-- Multics

As many of the other posters have pointed out, this work isn't necessarily new, but it is news.

There are other tools out there which do this: Legion, Avaki, Sun Grid Engine, Globus, to name a few but the goal is to create a network of (mostly) supercomputers which doesn't require a lot of reconfiguration at each site. What differentiates this work from many other approaches is that it is transparent to the system administrator.

For those who ask "why can't you just do something let seti@home" the answer is that not all problems in science and business can be easily decomposed into small chunks. Bandwidth requirements and latency may also be a problem. A lot of scientific programmers have to worry about communications much more than about processing power (although this tradeoff has been seesawing backwards and forwards with new advances in both technologies).

There's a worldwide effort through both business and academia to create a number of good, interoperating frameworks for doing this sort of transient, virtualised supercomputer.

Have a look at the Global Grid Forum (which is becoming the focus for Grid computing standards) for more information.