How The Game Is Played

Happy Birthday to Duke!

Tomorrow is the 10th anniversary of Java and I started looking back a bit at my own Java history.

I began at sun almost 6 years ago as the first full-time hire on the JDK Performance Tuning Team. Steve Wilson was my manager. But he wasn't my very first manager. In a typical Sun story, Steve was a contractor working for a guy named Dave (Sorry Dave I forgot your last name. I'm awful at names.) The week I was hired, Dave decided to leave. Steve interviewed and recommended me, and then a week later I recommended him as my replacement boss.

I think having the chance to effectively hire my own manager was a good introduction to Sun culture. In my 6 years at this place I've found it to be usually egalitarian, often surprising, and sometimes just a bit bizarre... but in a good way.

I am still VERY proud of the work we did on JDK1.3. JDK1.2 was a milestone as the first Java2 but it was a WHOLE lot of new code and there were some performance issues and bottlenecks that we, along with our volunteers from the JDK API teams (and there were more of them then us) worked out and got rid of in 1.3

Honestly, most of the heavy lifting was done by the API teams, our biggest contribution was creating processes for measuring performance daily and reporting problems and regressions immediately so they could be killed before they escaped.

The OTHER part of that job, and one that I loved, was Java performance programming evangelism and education. Through talks with some key Java customers of Sun's Steve and I quickly learned that half the performance problems out there were just programmer ignorance.

Through speeches at JavaOne, user's groups and SHARE meetings, as well as the book Steve and I eventually co-wrote (http://java.sun.com/docs/books/performance) , I had the joy of helping to re-educate a lot of Java programmers. We taught them the simple mistakes they could avoid, as well as how to profile and tune code. That book is still one of my prouder achievements at Sun. There is nothing better then to have a programmer you've never seen before come bounding up to you and say with a big grin, "I bought your book and it solved our performance problems!" There is a joy of pride in that that I am not ashamed to say I absolutely love!

I can't talk about teaching people to profile and tune their code without tipping my hat to the Java VM team. One of many brilliant things the VM guys have done over the years, was to develop a solid profiler interface early in Java's life-- JVMPI. This has made it one of the easiest environments I have worked in to analyze for performance bottlenecks and definitely helped us in helping Java shed its myth of slowness.

Be that as it may, its surprising how long those myths persisted. I guess a first impression is just as hard for a language to shed as it is for a person to shed. I would say it is only in the last year or two that I have finally stopped having to correct people's mis-perceptions of Java speed. I really believe a large part of Javafinally losing the label of slowness has been due to the work I have done with Chris Melissinos since leaving the performance team . Together for more then 3 years now, we have been encouraging Java game developers and getting their work showcased. When I left Steve's group I said to him, "Steve, when I can show Quake running in Java at the same speed as C, no ones ever going to call us slow again."

At least on the desktop, I think we are finally there.

I can't leave this subject though without mentioning the Sun Hotspot VM team one last time, and our counter-part to Steve's JDK group, the Hotspot VM performance team.

These guys are amazing. Consider this: They have less then 10% of the man-power IBM has committed to Java VM development. Nonetheless they make a VM that is at least IBMs equal and I think in a number of ways its superior. They give me amazing power to work with and for that I am eternally greatful.

My latest gift from them came when I was writing the Megacorps demo for GDC. This snuck in so silently that I don't know if anyone else has even noticed. The Megacorps port was a crash 3 week project. I had to write a whole fake networking layer to clue their code to my back-end and frankly didn't have time to write carefully optimized code. In particular I ended up having to allocate a LOT of short lived objects and copies of data. Breaking the rules I had preached to others about watching allocation of garbage in tight loops, I was scared to death I was introducing gc pauses.

When I finally ran it... it ran flawlessly. Sometime between the last time I tried this and now, the VM guys have gotten short-lived object allocation to where they always said they could-- its invisible!

So going into the parties and festivities tomorrow, there are a great many people who have worked very very hard on Java. And they all deserve praise and attention... but I'm going to be drinking to Sun's VM teams.

The finest in the world.

Posted at 01:05AM Apr 22, 2005 by gameguy in General  |  Comments[0]