Mark Baker writes...
I'm calling bullshit on WS-Transfer. Please join me.
"I have a mind like a steel... uh... thingy." Patrick Logan's weblog.
My favorite observation from Infoworld's Executive SOA Forum last Thursday came from The VP of Enterprise Architecture at Sony Pictures Entertainment. Pointing out that SOA is nothing new, he had first encountered the idea in Tandem's shared nothing messaging capability. (pdf)
The same observation was made by Alan Kay in the 1960's about the Burroughs B5000 and contributed to his invention of Smalltalk. And then there's Erlang and other shared nothing systems.
You wanna make fourteen dollars the hard way? (WAV)
-Rodney Dangerfield, Caddyshack
Tim Bray writes about WS-* vs. ReST/HTTP...
Speaking for myself, not for Sun, I think that we ought to be pouring resources and investment into tooling and developer support around simple XML/HTTP/REST technologies. You know, the standardized ones that work today.Vendors have sunk a lot of time and money into SOAP and WSDL. Appearances are this effort has gone into easing the burden of programming with SOAP and WSDL. That should tell us something about their languages as well as something about SOAP and WSDL.
What's the result?
Adam Tratchenberg spoke about ebay's approach to versioning their numerous web services at Infoworld's SOA Executive Forum last Thursday. When asked how ebay performs compatibility tests across multiple languages and toolkits, his response was they test with Java, .Net, and PHP. My guess is they don't test with many of the Java toolkits. Perhaps they test with .Net's various toolkits because Windows many programmers are in various stages of adoption.
Anyone with an incompatible toolkit or language though is not left out in the cold... ebay provides an HTTP and POX interface for each method. Can ebay's approach be considered a success for SOAP and WSDL given this lack of universality?
What if the vendors sunk as much money into making HTTP and POX as convenient as they're trying to do with SOAP and WSDL?
Meanwhile at least the long march toward more compatibility in WS-* continues. I've not found a list of attendees to Microsoft's interop session. My sense is the list of languages and toolkits is fairly short (Indigo and Sun's Java toolkit. Others?)
Is this a barrier to entry that guarantees its ultimate demise?
Why the fuck do people have to insist on complicating things?Maybe it is the case that S3 is the simplest thing that could work. But I think there is something between S3 and the full-blown web hosting provider he's fearing.
The simple way S3 works is as an associative memory between keys and values. As long as you know which key to use, this is a great service. But what if you need to retrieve your data based on some contents?
The key / value pair approach of S3 is the degenerate case of associative memory. I don't think S3 has to be about anything more than storage, but there I think there will be some room for associative retrieval of some complexity beyond simple keys.
If the pipe was big enough then there would be no trouble bringing gigabytes over the pipe and doing the associative lookup remotely. Or one could store in S3 the data itself under one key and various indexes under another set of keys. That's clunky but I can imagine someone trying it.
mdavidx5's point is well taken though. Keep it simple.
Stephen Williams writes in the FoRK email list on an exchange about Amazon's new storage service...
I firmly believe that both full filesystem semantics and ACID integrity constraints are red herrings and have seen a number projects reach the same conclusion...Having worked on and with very complicated object-oriented databases, very complicated relational databases, and other approached, I've noodled for some time on the possibilty that there are three basic persistence patterns that can meet a ton of needs.
With database-based applications, even when you have ACID capabilities, there are a number of reasons to avoid updates, avoid "accumulators", and otherwise avoid many of the situations where you needed transactions to begin with.
I will be interested to learn if Amazon has now or will offer some higher-level services on their storage service. Search, matching, versioning, etc. I see they want to keep it simple, and I agree with that. Before long though people will want to do things with their storage. Some of those things will be better done very close to the data itself.
I wonder if/when we'll see the ability to put computations very near Amazon's storage (including indexes, calculations, searches, etc.) that are aware of the format of the stored data, that is secure, etc. Storage is just the most basic start of a shared grid of services.
A problem with our industry (is it more than ours?) is the loose use of terms that truly have a formal definition. What is the meaning of this claim?
In contrast to standard distribution middleware such as CORBA or Java RMI, an SOA implements processes as first-class entities.This is on page 58 of the March/April IEEE Software magazine (pdf). I wonder if the editors have any more sense of how wrong this is than the authors?
First of all, SOA has no formal definition. Secondly, in any common use of the term SOA, there is nothing resembling a process, let alone a first-class process.
Smalltalk has something close to first-class processes. Kali Scheme, yes, even distributed first-class processes. Termite Scheme, uh-huh.
SOA is so far from having anything resembling first-class processes that such a claim in an institutional publication is incredibly disheartening. In 2006 our programming languages *should* have first-class processes.
Mark Baker points to a really interesting application of REST to dynamic networks of small devices.
The primary mean of abstraction within pREST is a resource. Virtually anything that can be uniquely addressed with a URL is a resource: devices, services, and configurations as well as documents, pictures, and raw data. Resources can be nested, so that services and properties of a component being associated with URL are resources themselves.
I will be at the SEM SIG meeting tomorrow night in Palo Alto to hear Mary Poppendieck talk about Lean Software Development (see quote below). Then Thursday I'll be at the Infoworld SOA conference in San Francisco.
Jon Udell, Phil Windley, and others will be at the conference, which should be worth the price of admission alone. Well, I got a free pass. So worth *more* than the price of admission!
I will be looking for entries to put on the ws-success wiki. The agenda looks good -- not just vendor fluff.
From Mary's abstract on Lean Software Development...
Lean Strategies for Software Development Leaders What do PatientKeeper, a hospital data management system, and Zara, a high fashion clothing chain, have in common with Dell Computer and Toyota Motor Corporation? All four companies are overwhelming their competition with a constant flood of new products that seem to be exactly what customers want, even as they set the standard for quality and value. How do they do it? To these companies, Lean Thinking is a way of life: Customer Value, Rapid Response, Constant Learning, Built-in Quality and Engaged Workers are part of the culture.
Jon Udell, et al. wonder about the value of open source and/or standards.
I would hesitate to adopt any standard that does not already have a up-to-date open source implementation. And I would refuse to define a standard without concurrently defining an open source implementation.
I've been involved in two standards definition processes. Neither had any significant implementation or formal model to guide them. I suspect other standardization efforts have suffered similar consequences.
Mistakes were made.