--- In [email protected], Patrick May
<[EMAIL PROTECTED]> wrote:
>
> On 27 Jun 2008, at 21:50, htshozawa wrote:
> > --- In [email protected], Patrick May
> > <pjm@> wrote:
> >>
> >> OpenSpaces uses the Spring
> >> dependency injection capabilities to allow business logic to run in
> > a
> >> Space Based Architecture without coupling the implementation of that
> >> logic to a particular vendor.
> >>
> > Interesting, but you're cutting out the major portion of the business
> > market because most commercial software from major software companies
> > are proprietary and do not use Spring.
> > An ESB based on Spring can bridge this gap by offering adapters to
> > proprietary systems (packaged and custom) while offering POJOs to
> > software based on open communities.
> 
>       This is exactly what GigaSpaces does, in terms of its Spring support.
> 
>       You're absolutely right about the number of COTS products that aren't  
> really designed for integration.  I've built a few systems that use a  
> grid (both data and compute) infrastructure to mediate the workflow  
> among such components.  One big advantage of the grid in these  
> environments is that it combines the advantages of queue-based and  
> publish/subscribe messaging without their disadvantages.  Messages  
> remain for as long as necessary and can be delivered reliably in  
> either one-to-one or broadcast mode.  This makes recovery from failure  
> much easier than with traditional ESBs because components can come up  
> in any order.  It also makes scalability easier because the grid acts  
> both as a distributed, in-memory shared state for stateless services  
> and as a deployment mechanism for those services.
> 
>       This architecture allows the development of highly resilient  
> distributed systems on top of unreliable hardware and software  
> components.  Because the workflow is mediated by the highly reliable  
> grid, the failure of individual agents (some of which use existing ESB  
> transformation tools) and backend systems does not stop the end-to-end  
> processing.  One example of this is an order management system I built  
> for a U.K. telco.  None of their backend systems (CRM, Billing,  
> Provisioning, etc.) were designed for integration.  They had (and  
> probably still have) a very flakey ERP system that once went down for  
> over half a day during peak season, bringing their old OMS to its  
> knees and costing the company over £2 million.
> 
>       One day a few weeks after the new system was in place, the head of  
> Web sales came running down to complain that "The OMS is broken!  We  
> haven't received any orders from the Web in 18 hours!"  I looked in  
> the grid and found a large number of orders waiting to be processed by  
> the, you guessed it, ERP system.  Once quick phone call to ask ops to  
> reboot the ERP app and we were back in business without losing a  
> single order.
> 
>       That's the kind of capability that businesses will pay for.

This recalls observations I have made in the distant past that rather
than expose Jini/JavaSpaces in the raw to unimaginative programmers,
it would have been more fruitful to dress the technology up in a form
with whose interfaces they were more philosophically familiar.  Mind
you, that was before middleware had sematically transmogrified into
that vague all-embracing entity known as an ESB...

Gervas

> 
> Regards,
> 
> Patrick       
> 
> ----
> [EMAIL PROTECTED]
> S P Engineering, Inc.
> Large scale, mission-critical, distributed OO systems design and  
> implementation.
> (C++, Java, Common Lisp, Jini, middleware, SOA)
>


Reply via email to