Rails in the Large: How Agility Allows Us to Build One Of the World's Biggest Rails Apps
Neal Ford shows what ThoughtWorks learned from scaling Rails development: infrastructure, testing, messaging, optimization, performance.
Tracking change and innovation in the enterprise software development community
Presented by Jürgen Höller on Aug 30, 2009
JBoss versus IBM WebSphere: Cost, Performance, Efficiency, Innovation (IBM wins)
Unix, Linux Uptime & Reliability Increase While Patch Management Woes Plague Windows (Yankee Group)
Comparing WebLogic, WebSphere, Oracle, and Open Source Application Servers
Lean development governance whitepaper by Scott Ambler and Per Kroll
What is he saying with "There is a very small percentage of systems that need remote transactions"? Is it true? Is the distributed transactions market over sold?
Mr. Höller sure knows what he is talking about - but, boy, is he boring. No jokes, no funny pictures, no funny stories - just slides with bullets - absolutely nothing entertaining. Yes, this is a technical presentation - but, if someone haven't notices, even technical presentation like this one are intended for people to consume. C'mon Jurgen - put some style in there - anything that can wake up some of the audience will do.
No jokes, no funny pictures, no funny stories
want edutainment?: try martin fowler. although not without substance, you won't find there the combination of breadth and depth of knowledge displayed here... and if you find a funny story about transaction management please do pass it on ;-)
cheers,
gerald
> Is the distributed transactions market over sold?
From my simple world view you can break distributed transactions into 2 cases.
1. You want 2 phase commit across multiple DBMS/JMS Queues
2. You want transaction semantics across network calls (EJB 1 & 2 etc)
The way I see it Juergen did a pretty good job of talking about the first case - aka 2pc over multiple resource managers has a significant cost and other alternative approaches exist.
EJB 1 and 2 is a good example of the other case imo. That is, it tried to make the network transparent (remote EJB calls) and pretty much needed a transaction manager to ensure those remote calls could all together be treated atomically (~ Corba anyone?).
In my book the common alternative to EJB 1/2 remote calls approach is to use web services. Generally each web service call executes atomically by itself - aka you generally don't have a 'transaction' spanning multiple web service calls (you might use state transitions of the underlying data instead).
That's my high level view of your question anyway. I end up converting your questions to...
- When is 2PC over multiple Resource Managers (DBMS/JMS) justified?
- When making RPC type calls when is EJB justified over say individually atomic web services.
Hopefully that is an interesting/useful perspective.
Cheers, Rob.
What is he saying with "There is a very small percentage of systems that need remote transactions"? Is it true? Is the distributed transactions market over sold?
You bet. It's always been. Most distributed systems are pretty much e-commerce sites or internal transaction processing systems, and in most of them (at least the ones I've seen), pretty much do local transactions.
Unfortunately, many of these systems set their connection pools to be XA connection pools, but they still end up using only one database, or if using multiple databases, their business rules get them to do transactions in each separately. It is very rare that you really have to set up a transaction to be committed as all-or-nothing simultaneously on more than one resource.
Blame it to software vendors and academia for forcing a theoretical white elephant that's only good for a narrow problem domain to the general IT populace :)
Yes, I agree that 2PC has been 'oversold' (in the JEE world).
Vendors do tend to skip out the "... you don't want to use this feature in this scenario because...". IMO this is why presentations like this are important.
For me part of the 'problem' was that JTA it was a requirement of EJB v1 due to its remoting/distributed nature... so by default people would tend to use XA/2PC for their DB transactions. So 2PC wasn't always *sold* but came with the J2EE territory. I'm not sure how many developers/engineers thought critically of the implications but instead went with the flow.
Neal Ford shows what ThoughtWorks learned from scaling Rails development: infrastructure, testing, messaging, optimization, performance.
Stuart Halloway discusses Clojure and functional programing on the JVM in depth, and touches on the uses of a number of other modern JVM languages including JRuby, Groovy, Scala and Haskell.
Oren Teich and Blake Mizerany talk about the technology behind Heroku and the benefits of the new add-on system.
Chris Riley presents security issues threatening service based systems, examining security threats, presenting measures to reduce the risks, and mentioning available security frameworks.
This talk investigates technical issues encountered when moving to an Agile process.
Don Box and Amanda Laucher present “M”, a declarative language for building data models, domain models or external DSLs. Don Box's demos show some of M’s features and latest changes of the language.
It is four months since the SOA manifesto was announced; InfoQ interviewed the original author’s to get insight into the motivations and the process behind the initiative.
This article explains the impact memory barriers, or fences, have on the determinism of multi-threaded programs.
6 comments
Watch Thread Reply