QCon London 2011

Once again I went to London, and once again it did not disappoint :) this is by far the best meeting of the minds that I have the pleasure of going to. Although I must admit I find the choice in flavours of cake a bit weird, chocolate and lime ?!?!

I heard some interesting talks, I wish that I had the confidence to ask questions when I am there as there are many people I would like to pick the brains of. Maybe next time I'll pluck up the courage to ask why "facts" displayed by two different talks were in complete opposition or why the facebook team seem to think the problems they have faced are unique to them.

So on with my break down of some talks I personally attended, the following is a series of opinions. The thing about opinions is they are mine and you almost certainly won't agree 100%.

Wednesday

Why Don't We Learn?

This was an enlightening, informal chat with the audience that explained the principles of learning combining many ideas from the psychology field rather than IT. The speaker Russ Miles was enthusiastic and fluid not having to read from slides and making sure that his voice projected. A brief mention of the placebo effect was the only item that wasn't accompanied by facts, a simple read of some Ben Goldacre could solve that.

Software quality - you know it when you see it

Once again a very practised talk that conveyed in a short time the fact that you really do know good code when you see it, or rather you can tell when there is bad code. Erik Dörnenburg produced a slick series of graphical views of code that can make a series of issues clearly visible even to inexperienced personnel that don't code for a living. Such things that we all know are bad such as monolithic classes a high degree of coupling as well as pointless interfaces.

Learning and perverse incentives: "The Evil Hat"

This was one of those choices that I was unsure of at the time, I needn't have been. The speaker Liz Keogh was a brilliant speaker, she mixed up real life experience and sarcasm just the way I like it. The actual content of the talk was essentially carrot works better than stick and if you set a kpi prepare to have it gamed! She cited one example of a kpi being gamed that rang true with circumstances that I've had the misfortune of coming in contact with. One of her friends said he had a new project manager that had told him that the project was to be listed as "in trouble" and that he was calling an emergency meeting. The project was meeting deadlines and bugs were minimal so he couldn't understand why it needed saving. Liz asked a simple question, "are pm's bonuses related to rescued projects", bingo, the pm would get a bonus for simply saying a project was in trouble so that he could be seen to swoop in and "save" it.

Agile Operations - optimising the business on shell script at a time

I am going to be slightly controversial on this one, so be warned. Dan as always was a brilliant speaker as was his co-host (I resist the term side-kick) Chris Read. However the talk didn't really do it for me, the comedy was good but over-played. I will admit that it's probably just me that doesn't yet get it as Dan North himself admits he didn't "get it" straight away. Yes it is nice to be able to deploy an image across multiple servers at the press of a button and send changes out in a similar fashion, but this doesn't seem new to me as it seems obvious.

Java without GC Pauses

Gil Tene was a comfortable speaker, I did get the impression from his demeanour on stage that he was beginning to get tired of having to explain the same things over and over again and people not getting it. I have to admit that this is the talk I had wanted to see the most all day, I like the clever stuff that happens at a low level thus my love of the JIT e.t.c.
The solution to the compaction problem seems so simple once explained, it uses virtualisation so that rather than change physical memory during compaction (the long part of a full GC) you alter the vector table of memory. The only issue that this has is that Azul's GC (C4 I think) does require being installed in unison with some virtualisation server...as I understand it anyway??? This idea seemed so obvious and if the charts are to be believed it is very effective so it seems a no brainer that virtual-pointers become the next change for the linux kernel. Loved this talk immensely, however on day 3 of QCon it was bought into question.

Thursday

Testing for the unexpected

This talk given by Ulf Wiger was a surprise for me, in a pleasant way. To be honest I had no idea which presentation took my fancy and as it is always good to learn something "totally" different from normal I went along. I didn't really know how we possibly could test what was not expected and in walk quicktest. Essentially you fill a series of tests up that are then filled with essentially random data. I tend to do this as a matter of course so the fact that there is a framework for it probably means I am missing a trick.

Data Architecture at Twitter scale - Nick Kallen

I can't really say much but it was a good talk, no surprises also no exceptional solutions to problems. The fact that I've written less about this talk doesn't mean it was less brilliant than the others it was great maybe I'll add some details later :)

Releasing Fast Code: the DevOps Approach to performance

I can't remember this one, hmmm need to review my notes

Better is Better

This was a somewhat strange talk. It was billed as being a bit of a rant and to be fair to the speaker that is exactly what it was. I did feel a bit sorry for the speaker as he was new to public speaking and had fallen into the same traps that all new comers do, slightly mumbling and stumbling. Things weren't helped by the fact that he moved off mic as he paced on stage. All this is rather negative at the end of the day it was his personal experience as a story, I found this quite interesting as I do enjoy hearing about the history of computing. The history of computing is very important at the moment as we seem to be re-living it, I'll blog what I mean on this later.

I went to another talk on Thursday but can't find the details for it unfortunately.

Friday

High performance web applications in Haskell - Steve Vinoski

I had high hopes for this talk, sadly totally unfounded. The speaker was very arrogant and childish, never passing up an opportunity to bad mouth another language. These "my language is better than yours so there nerr nerr nee nerr nerr" talks drive me mad. I feel like grabbing the guy shaking him and saying "Ok I get it you hate anything that isn't your language, tell me why yours is good not why any other is bad." these people say this language is more concise without realising that there is context to this claim for example recursion makes for very concise code in a lot of cases but not all plus there are other concerns to a language readability, maintainability, performance (that's right sometimes functional is not faster than procedural sometimes it is), plus there is always the expertise of the people looking at the code.

Anyway I digress this guy got off to a bad start, he did reign it in once he concentrated on the idea of the talk, however I waited to hear some Haskell my functional language of choice only not to hear any he did mention Erlang and throw some (badly formatted) slides containing very basic Haskell up but there was no depth to it. I actually tweeted a bad thing which I don't like doing as concentrating on the negative is not good, unless it's funny :)

HTML 5 Design/Development Tooling (+HTML and Flash)

Not really a talk this one, more of a sales pitch from the Adobe guys. I liked seeing what can be done but would have liked to have seen what was going on underneath the tools if that makes sense. I'll admit I am no longer into web design and GUI's as I find the server side more interesting so have fallen behind with HTML5 seeing what can be done in this talk has prompted me to learn about it especially as the company I work for use flex to produce their GUI. My work colleague who joined me for this presentation was not interested in the talk as all of the tools being demo'd he had already seen so sorry @DavidArno but I did enjoy it to a point. There is however the danger that QCon talks will start to become more about selling stuff than pure knowledge sharing which would be a terrible thing to happen. On that note there seemed to be more talks around a product (twitter, facebook, dynatrace, atlassian, adobe, azul ...) than last year so maybe the knowledge rot has already started.

Performance Tuning for Java Applications

First of all this room was far too small for the number of people that attended the talk, it got really hot and at one point I had to leave to prevent myself passing out only to return to find someone sitting in my seat which I had left my belongings on to inform anyone that I was going to return, some people are dumb.
The speaker was great, it felt like an informal chat about how he tests products at Atlassian I don't know if it is a cultural thing but most Australians and Australian companies seem to foster a very easy going style that I respond well to George Barnett spends his time at Atlassian performance testing and really knows his stuff. I was however concerned at one point though as the charts he showed for heap size were in complete contradiction of the ones the Azul guy showed on the first day. George said if you have the memory available then use it in the heap and showed the performance increase even when the complete heap wasn't being used, whereas Gil Tene showed that compaction increased in stop-the-world length of time to run which had a marked effect on system performance. The only thing I can think that could make them both right is that the applications they used for testing were utilizing the heap in very different ways. confluence and jira for example should have short lived classes of small size so that even if they live long enough to reach old space the gc could easily remove them with concurrent mark sweep and not have much compaction whereas Gil may have used a system with irregular classes and life expectancies. This is something that George made clear at the start of the talk he was ONLY using Atlassian products and that anyone who blindly applied his performance tweaks to their system deserved every bit of failure they got, unless the system they were tuning was an Atlassian product.
Following on from this I did learn that in order to performance tune an application you have to know the application, how it uses memory and indeed what the current performance is before making any changes otherwise how do you know if you are making any improvements.

Using a Graph database to power the web [Rick Bullota & Emil Eifrem]

I did really enjoy this talk, it didn't expose me to anything new but was nice to hear that a few areas of mathematics that I enjoy are bubbling up to the surface graph theory and set algebra. The talk could have been more weighted to neo4j than thingworx as I was interested in the technology used more than they was a single solution was applied, but case studies are a way to learn I suppose as long as you remember each situation carries it's own unique attributes.

HTML5 and the dawn of rich mobile web applications

I didn't hold out much hope for another html5 talk given the sales pitch Adobe talk from earlier, but boy was I wrong.
This guy (James Pearce) was so enthusiastic and enthralling that you couldn't help but fall in love with the subject matter every bit as much as he obviously had. He started by demo'ing how we have learnt that the web is not just sitting at a desk and that it isn't just a change in screen size that we need to take into account but the whole interface and what visitors to sites want to do. He cited some good and some unexpectedly bad examples of where this is not done before showing some simple code and explaining more about offline html applications. This was a fast paced exciting talk I was blown away especially as it was a subject that over the past few years I've shown less interest in.

#qconlondon