Blogarithms
Doug Kaye's thoughts on web services, web hosting and managed services.
|
IT Conversation with John McDowall, CTO of Grand Central Communications. It's the audio version of John's top-ten list of Rules of Service Design:
- Design services to be shared.
- Services have a clear purpose.
- Services are discoverable and support introspection.
- Services plug into a SOA.
- Services can be loosely orchestrated and use other services whenever possible for common tasks.
- A service has a well-defined use policy/contract.
- Services accept well-defined input and deliver well-defined output.
- Services do not have hidden side effects (play well with others).
- Services are interfaces to or from processes.
- Services must provide visibility and an SLA.
[stream—download—discuss, 5.1 mb, 22 minutes, recorded 7/30/03]
Posted Thursday, August 14, 2003 10:59:29 PM 
|
|
|
The Grid: Tightly Coupled. My sister called via cell phone from Manhattan this evening. (She had nothing better to do in the dark.) She said, "You're so lucky. You have solar power, so a blackout like this wouldn't affect you."
Not true. Ours is an on-grid photovoltaic system. We're a cogenerator, just like any other, and when the signal from the grid goes away, our system drops off the grid and shuts down. We *could* rewire our system so that we could isolate from the grid and use our own power locally, but it would have to be on dedicated circuits only, whereas the on-grid system just adds our juice to the big pool out there. And why do we, like all other generators, have to leave the grid when it dies? One reason is phase lock. We can't put electricity onto the grid unless it's in phase with all the other generators on the grid. For our little 2.5kw system, it's just a matter of a solid-state inverter controlled by a phase-locked loop. But if you're running a huge mechanical turbine (i.e. with mass), it can take a long time to get it to the correct speed and phase, and when you suddenly put a load on it, it starts to slow, which means your frequency drops, you phase-lock goes to hell, and...well, you get the idea.
I wish I still had an article I read in IEEE Spectrum that described the step-by-step calamity that lead to the huge NY blackout in 1977. It was written like an action thriller with timestamps on each paragraph. I'll let you know if I find it online.
Posted Thursday, August 14, 2003 7:47:33 PM
|
|
|
Gartner on Event-Driven Architectures. This is an excellent web seminar by Yefin Natis and Roy Schulte of Gartner in which they explain the concepts of asynchronous event-driven services. In his review of the presentation, Edwin Khodabakchian at Collaxa pointed out that the Gartner distinction between SOA and EDA may not be correct. Indeed, Gartner seems to have decided that the phrase "service-oriented architecture" can only apply to the request/response RPC-style synchronous MEP (message-exchange pattern). Most experts consider all MEPs, whether sync or async, to be under the SOA umbrella. But it's a detail, and the presentation is valuable nonetheless.
The total presentation runs an hour, but if you don't need the introduction to the basics of SOAs, skip to 27:00 and just listen to Schulte's half. (Make sure you get the right presentation. There may be more than one on the page.) That's where the event-driven elements are presented. I particularly liked Schulte's point that while synchronous systems involve a single reqeustor and responder, an event-driven asynch system publishes events which have a potentially unlimited number of subscribers (as in JMS). As he says, the flow is determined by the recipient, not the originator.
Posted Thursday, August 14, 2003 7:16:06 PM
|
|
|
MSDN TV: Loosely Coupled Web Services Software legend-to-be Doug Purdy explains how to create .NET web services that are version resilient using the Open Content Model. Specifically, Doug shows how to write clients that can handle unknown versions (backward-compatible) without the need to recompile. It's a real-world example of loose coupling. [32mb WMV file]
Update: Radovan Janecek, VP engineering at Systinet, has a new blog that will focus on web services. (Welcome, Radovan!) He he points out that the technique Purdy demonstrates to obtain version resiliency depends on dumping structured data into an unstructured and untyped string. He's right: it's a bit of a quick-and-dirty gimmick. Still, I like the video because it demonstrates the problem if not the most elegant solution.
Posted Thursday, August 14, 2003 1:11:32 PM
|
|
|
|
|
|
