Gregor's Ramblings

SATURN Conference 2015

May 1, 2015

Gregor Hohpe
Hi, I am Gregor Hohpe, co-author of the book Enterprise Integration Patterns. I like to work on and write about asynchronous messaging systems, service-oriented architectures, and all sorts of enterprise computing and architecture topics. I am also the Chief Architect at Allianz SE, one of the largest insurance companies in the world.

I have not blogged about events in a while (my Rambling category tells me almost 6 years!), but SATURN 2015 in Baltimore has been an amazing event that's well worth rambling about.

The perfect blend

One interesting aspect of SATURN is that academics call it an industry conference whereas folks from industry tend to call it an academic conference. While this can lead to minor annoyances (yes, it is an industry conference), I feel it highlights why the event is so unique and useful: it is a perfect blend of structured thinking from the academic edge combined with valuable industry experience that would otherwise not make it to an academic conference due to lack of analysis and validation. The blend worked extremely well for me. It also implies that there is no product or vendor pitching at this event, aside from the occasional book pitch by the author (which should be acceptable - you would do the same after going through all the effort) and brief mentions of the sponsors, including Allianz.

Celebrity Spotting

Len BassIf OOPSLA used to be the spot to hang out with the "big boys" (and some girls), SATURN has likely taken its place. If you were going to bring all books you could get signed from the folks attending the event, you'd need to check an extra piece of luggage: Mary Shaw, Rebecca Wirfs-Brock, Len Bass (his books alone would have you pay an overweight charge), Simon Brown, George Fairbanks, Cesare Pautasso, Doug Schmidt, Joe Yoder, Olaf Zimmermann. And I am sure to have forgotten some folks.

Microservices Workshop

Monday was an all-day microservices workshop setup by George Fairbanks and Michael Keeling. One may feel that there is not much left to say about microservices with Sam's Book out and Martin bliki-ing about it. Still, thanks to the superb moderation and the caliber of attendants I found the workshop to be very useful. We split into subgroups, of which I attended the one on "API Design", always a topic of interest. We discussed way too many things to even fit into a rambling, so here just some sound bites:

  • Interaction Models: API design focusses too much on data representation and too little on interaction models. I am trying to help solve this.
  • Domain-driven Service Design: For the data representation aspect and the "slicing" of micro services we should have a fresh look at Eric Evans' Domain-driven Design. Much of his guidance should apply to service design as well. For example, domain boundaries might make good service boundaries.
  • Versioning: Service API's should be immutable, relegating evolution to new versions. Another option is to do what Google did with Stubby and ProtoBufs's: they push the versioning problem into the application layer by extending existing data formats with optional fields and designing services such that they can accept calls from "old" callers who do not yet supply the new fields. In the easiest case this is done with defaults, but it can require the service to check which fields are present and to execute the corresponding version of the service. It's clever, but from my experience it depends on thorough documentation of the behavior inside the proto definition files.
  • George moderating microservices workshopIt's all connected: While this sounds like the dream of a person enamored with integration, at the logical level things are indeed also interconnected: the current complex systems and rapid cycles have too many moving parts and too many client versions to rely on complete testing upfront. Therefore you can only test in production, e.g. with canaries (deploying new versions to a limited number of servers to see how things go). This raises the question whether users are happy if they experience different versions of a service or web site, sometimes even ahead of announcement. If your changes are small and incremental, it works relatively well. You can have small changes only if you have short release cycles -- otherwise you are making no progress. This implies that out of the Micro-services / Agile / DevOps mix you cannot arbitrarily pick a subset - its "all in or nothing". Martin makes a similar argument in his You must be this tall to use microservices article.
  • Governance: Do you let 1000 flowers bloom or do you exercise central control? To me this sounds like a classic decision analysis question: If you decide upfront and exercise governance you have lower cost -- as long as you guess right. If you guess wrong you face high rework costs. Letting things evolve over time has a higher inherent cost, but also a higher chance of leading to the "right" answer.
    Governance also takes many more forms than a fat book and a big hammer. For example Netflix (all good exampels these days seem to come from Netflix) provides a client library that include circuit-breaker functionality. Of course not everyone uses that library. So they run little disturbances in production that causes your service to get hung -- this helps adoption!
  • Dynamic properties: Mary Shaw reasoned that systems using micro services architectures are emergent, meaning you cannot reason about their behavior until they are assembled and running. She would place them in the "implicit invocation" block of architectural styles because the invocation of one method by another is not explicit like in OO systems, but implicit like in pub-sub systems.
  • Monitoring: In such a dynamic system, monitoring is paramount. Len mentioned that Netflix collects 90 metrics for their canaries.

Mary Shaw Keynote

In case you did not know, Mary Shaw co-authored the first and seminal book on Software Architecture with Davie Garlan. A key aspect of the book is the categorization into "architectural styles". For example, we cite the Pipes-and-filters style in EIP. She described that the concept emerged when she collected a large number of technical documents which tended to include roughly 6 column-inches on "the architecture of my system" supplemented by about 2 column-inches of diagrams. Looking at a large number of these sections, certain categories emerged, which after a while solidified into architectural styles. Harvesting design patterns is actually a quite similar process of looking for common approaches, although the nuances are somewhat different: patterns dig deeper into the forces and intents of the recurring solutions.

Mary Shaw KeynoteMary (unsurprisingly) talked about software engineering, with the core thesis whether software engineering does in fact deserve to be called engineering. She drew on civil engineering, specifically bridges (Cindy would be happy). It's actually quite amazing that the methods devised by the Roman's without any theory and very little mathematics lasted into the 19th century. A key element to their success was empirical progress: they learned from failure and kept refining their concepts. So real-life buildings can be agile, at least if you have enough time and cheap labor. If you want to add a true classic to your bookshelf, get Vitruvius' De Architectura from 25 BC. The science of bridge engineering came much later, e.g in the form of tabulation of iron properties in the mid 1750's. A hundred years later engineers were able to do a full analysis of a bridge, at least minus resonance.

So how does software engineering compare? In oder to start from the right mental model we have to remember that software is design intensive with negligible manufacturing cost: coding is design. Manufacturing used to consist of printing CD's but now consists mostly of uploading binaries to a distribution site. People often feel that there is no science software development or software engineering. Mary showed that there is quite some science in software; it just takes 15-20 years from basic research to popularization, at which point it is no longer called "science". In our world much basic research happened in the mid-80ies, e.g. the notion that connectors are as important as components. I guess it's time I finish my interaction patterns... The early 90ies saw concept formation followed by much development, exemplified in many conference tracks on software architecture and engineering. Internal exploration brought focus on formal analysis, quality attributes, and methods such as ATAM. External exploration brought us processes such as RUP and component-based software engineering. Popularization was reached in the last 15 years with many standard interfaces and frameworks, many of them driven by the evolution of the Web.

But software engineering still falls short in many aspects. "I am trying as hard as I can" is not engineering, Mary reminded us. We fall extremely short in making reference materials available. Every chemical engineer can reach for his copy of Perry's Handbook -- we have no such thing. Even Grady's effort on the grand Handbook of Architecture seem to have stalled. One key difference is that Perry's Handbook has a 15 year update cycle, something that would not do us much good (side note: it looks like EIP is heading for a similar update cycle). Search has become our method of sharing knowledge. While this is a fact of life we should consider whether crawling discussion groups is really the path to more mature and reliable software engineering.

Sam Newman on Microservices

It was great to see Sam again and reminisce about old stories including pasta tongues imported from Japan. He gace a great overview of what it means to use microservices. Rather than rehash it here I recommend you read his new book.

Later Simon and Len put Sam on the "Microservices trial", which yielded a ton of great sound bites and pearls of wisdom, including:

  • Box-box-cylinder as architectural style (Simon)
  • Monolith vs. microservices style is like America vs. Europe (Les)
  • Using the network enforces discipline in setting boundaries (Sam)
  • How would you use a chaos monkey with a monolith? You'd have to take the whole system down! (Sam)
  • We should be cautious with standards. UDDI was a "lets solve a problem we have not solved yet" type of standard (Sam)
  • Tool progress has allowed us to go smaller, e.g. automation, containers (Sam)
  • If you have a brand-new product, first build a monolith, because you won't know where the seams are. Then go to microservices later. (Sam)

Software Architecture as Code

Simon Brown and I bonded on software visualization, especially aided by annotations. I should visit Jersey and he should talk with Erik Doernenburg.

80 Years of Patterns Impact

Gregor, Rebecca, Joe, OlafSATURN was also a great moment to bring together the 4 authors of our 20 years of patterns' impact paper: Rebecca, Joe, Olaf and myself. I am still entertained by the thought what the odds are that you can author a paper with 3 co-authors whose names start with W (Wirfs-Brock), Y (Yoder), and Z (Zimmermann). Bobby should talk to some of these guys :-) The 80 years are maybe a slight exaggeration as for example I was not part of PLoP until 2002 when we workshopped EIP. The picture shows us in correct alphabetical order.

Thanks to all for a great event. There were many more great talks and conversations (and some Belgian style brews) that I did not manage to write up. And we stayed safe despite the ongoing protests in Baltimore. In the end we only saw a single and very peaceful protester on the way home.

Last, but certainly not least, I should thank my muse for inviting me to SATURN!


Follow:       Subscribe  SUBSCRIBE TO FEED


Gregor is the Chief IT Architect of Allianz SE. He is a frequent speaker on asynchronous messaging and service-oriented architectures and co-authored Enterprise Integration Patterns (Addison-Wesley). His mission is to make integration and distributed system development easier by harvesting common patterns and best practices from many different technologies.