Archive for the 'Distributed' Category
Subjugating Temporal anomalies to satisfy software’s Prime Directive
Having just read one of Howard Dierking’s latest posts regarding how we reason about time when it comes to software, I am reminded about the benefits that one way messaging and publish / subscribe can bring to the table. It stands to reason that we all want our code to execute faster and our applications to respond to their users in a timely fashion; ‘slow’ is not a word users or developers alike want to hear when discussing or describing their systems. Howard’s post is a discussion that came out of his reading a book by Steven Seow titled “Designing and Engineering Time: The Psychology of Time Perception in Software”. I personally haven’t read the book however my interest in doing so has certainly been raised. I am interested here in turning the architectural spotlight onto the strategies for managing time.
Howard includes the following table in his post (the first two columns), demonstrating how perceptions of time in software / human interaction can be managed effectively. As my mind turned to thinking about publish subscribe and one way messaging I decided to add another column to this table clarify my thinking.
| Performance Dimension | Management Strategy | One Way Messaging Pub / Sub Strategy |
| Actual Performance – the precise time durations for completing a task | Operational Management – purely technical strategy (e.g. change the way it works to make it faster) | Let’s push aside algorithmic tweaking designed to speed up computation for just a moment and consider this aspect from a scaling / throughput perspective. In systems that attempt to address this through stateless web services, the strategy is one of scaling out the web servers. If we bring durable queuing into the focus then we can rather more cheaply apply divide and conquer, with patterns such as ‘competing consumer’. In terms of NServiceBus the Distributor is an excellent choice here. Again this is looking at it from an architectural viewpoint. |
| Perceived Performance – the approximate duration for completing a task as experienced by the user | Perception Management – make the task feel like it takes less time (examples include distractions, progress indicators, etc.) | Sending commands or messages to the server for processing and distract the user by maintaining UI state as though the command had been processed. One example is showing blog comments to the submitting user only, whilst waiting for the moderator to approve it. |
| User Tolerance – based on expectations, the approximate duration threshold where anything beyond will be considered slow by a user | Tolerance Management – focus on making users more tolerant of a longer duration when it cannot be shortened or “disguised” | Initial immediate response followed by a subsequent response via a different channel than that sent the message / request. Examples such as Amazon purchases that provide an immediate response and then provides confirmation email. |
In request / response systems, often the focus goes into that described in row 1 column 2 above; that is we tend to narrow in on speeding up our code, substituting for faster implementation technologies or even scaling up our hardware. In the case of stateless Web Services it’s also common to scale out the web servers. With one way messaging and NServiceBus we can increase our throughput and speed (both real and perceived) by scaling out our services and what’s more this can be achieved relatively cheaply with commodity hardware.
It’s also worth noting that the effects of implementing durable queues play’s into the Reliability aspect of the system and when those queues are placed physically adjacent with task based smart client applications that both send and receive messages, then it’s possible to create the perception that work is submitted immediately; potentially it may be repudiated later if there is something wrong with the submission itself. Maintaining a low number of these repudiated work submissions (command / messages) can increase the tolerance levels and the perceived improvement in speed in getting through the work itself.
Oslo, SOA and BizTalk - Speaking from the inside out now!
I am spending more and more time with my head wrapped around BizTalk in one way or another and very recently I am glad to say that I successfully navigated the MCTS exam and certified as a BizTalk specialist. Does this mean that I expect to do a lot of work in BizTalk from this point on? Good question. Well I do expect to do some yes and I am hoping that I will be able to offer the measured position of being from the camp that does not reach for BizTalk to solve every SOA problem looking for a solution.
There are alternatives
I do appreciate many of the features of BTS 2006 but I must confess that sometimes I do hear some absolutely mad assertions coming from those who would throw BizTalk at every problem where messaging was a clear choice. I have heard 
voices hold forth that ALL business rules should be managed in one central location (in BizTalk) and that all applications and line of business applications should make BizTalk their messaging target endpoint. I remember objecting to this idea firstly on the grounds that it was out of alignment with the budget constraints of a given project and secondly that it was clearly propagating fallacy number eleven of the fallacies of distributed computing. Now in this particular case my argument was pretty solid but I should also point out that it’s not always that cut and dry. BizTalk may well be the right architectural and financial choice in many cases where we know that scaling out on mass wont be a future consideration coupled with a short delivery window imperative or perhaps a business lacking in the capability to maintain a considerable software team required to maintain a complex custom SOA application. This does not exhaust the list no doubt but gives an idea.
SOA without the bells and whistles
I may want to build my own very light weight SOA framework or use NServiceBus, either option affording me the option to deploy multiple versions of the same service at different geographical points across my LAN. In this scenario my rules should stay with my service, I don’t fancy the idea of wearing the expense of BTS for every long running (or otherwise) service I write and I have all those nicely designed domain entities that have come out of my having been Domain Driven all along the way! Heck I may even be using WCF and WF in the most out of the box fashion and even then it makes very little sense to put all my business rules in one place just because I can. Centralizing things in my architecture expose me to a single point of failure, which we know isn’t a great idea. Some will argue that it’s easier to maintain in one place and if our environment is lacking in complexity then there are times where that does hold true, but over time as developers and architects begin to pile things into one place things might start to unravel. Imagine a BizTalk environment that begins to exhibit as a massive repository for your organisations business rules! Hmm, if I change this one rule here what happens to all the logic that uses my other 1300 rules and which one of them have dependencies on the rule I just changed? I should clarify here: I am not ready to throw out the baby with the bath water just yet however, I wouldn’t even begin to consider for a moment the notion that I should build tools for mapping or adapters for some of the LOB applications out there, these very mature aspects of BizTalk can exist very nicely behind my services rather at the front of my design.
Figure 1.0
Oslo is certainly suggesting architectural decisions like that exemplified above, where the BizTalk hosting environment continue to be used for integration services and the ‘Dublin’ Server be used for application services. This distinction has me curious; will Oslo’s process server be similar to BizTalk with respect to being a Broker? In the meantime as I become more adept at using the product, I feel more at peace with making up my mind about why, when and where BizTalk makes sense.
Oslo, SOA, BizTalk Express and crossing the chasm (part 4).
BizTalk Express - Oh well got the name wrong - they called it Dublin. I want to
make this the last post where I use the term BizTalk Express.
PDC is on our doorstep and the last minute information leak is starting to turn to a steady stream, the one that most interested me recently came from Darren Jefford just a few days ago. Darren speaks about how Dublin (the new WCF / WF Application Server Host) will help in making both those technologies scale to enterprise without developers having to write a whole lot of infrastructure code. Darren say’s….
“So if you want to expose [WF] workflow’s via [WCF] services but ensure performance and scalability (up to enterprise scale), you can now do this without having to write the code required to host these apps on Windows Server. Ensuring performance and scale of WCF services and WF is hard to do today, hence it’s not done very often at least in my experience and sometimes causes a tendency to twist BizTalk into doing something it wasn’t necessarily designed to do which causes problems of their own (coupling Web Sites/UI’s directly to BizTalk for synchronous processing springs to mind).
We don’t want customers in this situation to be forced into writing huge amounts of hard plumbing code to achieve this, we need a server product to do this for you, which is where Dublin comes in. Note some of the server features announced which will be familiar to BizTalk developers (content based routing, compensation, etc.).”
So it would appear that some of the benefits of BizTalk are being borrowed and leveraged to provide application developers the ‘Server’ or ‘Host’ product and even some content based routing (both requests in the call for BizTalk Express). However part of my previous assertions was that by providing BizTalk Express (conceptually), we might become the beneficiaries of a robust and scalable host for our messaging and long running workflow technologies, there is still one point that I maintain is of significant importance that should not be overlooked and that is the requirement for Publish / Subscribe. So I am still left with this question dangling:
“How deep will all this declarative tooling go? How far does Dublin reach down into the API? Do I get publish / subscribe with my WCF / WF messages / workflow’s? What about durable messaging regardless of transport and bindings - I don’t care to lose messages on the HTTP stack?”
Whilst the information seems to be coming a little thicker and faster, some of these 
questions remain unanswered. I am also very interested in how flexible these new shiny toys will be? I for one don’t particularly like the RPC / CRUD styled interfaces that you see being developed ad nauseam in the Web Service space, that approach does not feel at all like messaging to me - it’s more like remote RPC API calls than it is messaging. Does Dublin and the Oslo suite of tools offer me the environment to take advantage of sending business messages or will it continue to encourage RPC styled API calls with better hosting and scale features out of the box?
I’m sure there is a lot more to the Oslo story that will be unveiled in the next month and I remain hopeful that some of my wish list is on the menu.
