The other side of the moon

So I was supposed to speak at Velocity 2010 this year. My talk was titled Latency: Why you should worry and what you can do about it. It was one of the last few talks at the conference. Unfortunately though, the thing about a conference like velocity is that almost everything is about latency, and by the time I went on stage, absolutely every point that I'd planned on making had already been covered.

That's not a bad thing. For one, it validates my ideas, telling me that many other smart people have been thinking about the same things. Secondly, each of these speakers had 60-90 minutes to convey their ideas, which meant they could really do it justice. It also shows that velocity is a pretty awesome conference to be at.

So, by the night of the 23rd, I realised that there was no way I could go ahead with my talk in its current state, and sat down to rewrite it. That's when the drama started.

Now my usual format is to build my slides using LaTeX and generate a PDF from that. I have a bunch of relevant images and charts in there, and it's not too hard to do that with LaTeX. I started going down that path, but realised that I wanted to do something different. I really wanted to reference all the other presentations and go over points that they'd covered, so I switched to Comic Life. I was up until 5am, working out the details and getting the script right.

The thing about a 12 minute talk is that you can't waste a single minute thinking about what you're going to say next. You just need to know it, and there wasn't much time for me to prepare. I'd also planned to announce boomerang in the latter part of my talk, so had to fit that in as well.

At 5 I realised that I was too sleepy to keep going, so I went to bed and figured I could finish things off in a couple of hours once I woke up. I was also too sleepy to backup the presentation to my server.

I woke up at 10, and opened up the mac to start work again, and it wouldn't come on. I could hear the fans whirring inside, and the startup sound would sound, but the screen wouldn't turn on. It was now officially time to panic.

Two things were in my favour though. I work at Yahoo!, and the conference was just down the road from Yahoo! HQ. I headed in to the office and took the mac down to IT. The guys there were really helpful. They opened up a new Mac, and transferred my hard disk to it. The new one started up perfectly, and it's what I'm using right now.

I spent the next two hours putting the finishing touches on, and adding details for the boomerang release, and then rushed off to the conference with a slice of pizza in my hand.

Once I got on stage, I plugged in the mac, and as it has done on numerous occasions in the past, it hung as soon as the VGA adaptor was plugged in. I had to reboot while on stage, and this was all coming out of the 12 minutes that I had.

The talk went off okay. Looking back at the video, there's a bunch of things I'd have done differently, and some more details that I'd have covered, but I'll let you decide for yourselves.

My slides are on slideshare, while the video is on blip.tv.

I've been playing around with comic life, and this is what I ended up with:


I'm calling it 4.01 Strict.

I've been silent on this blog for a while, and the main reason for that is boomerang. It's taken a few months to get it ready for a public release, and I'm quite pleased with the way it's come out.

As many of you know, I've worked with the Exceptional Performance team at Yahoo! for over two years, and during this time, I've done a lot of work with measuring and analysing the performance of all of Yahoo!'s websites (yes, all). There are several tens of thousands of page types, and probably hundreds of thousands of unique URLs that could show up on a yahoo.com domain. This throws up some very interesting problems that I'm not sure I'd have had the chance to work with anywhere else.

I learnt a lot about scaling cheap hardware to meet the needs of this system, and also a lot about how each of these thousands of pages worked to improve their performance.

The measurement tool that we use has turned into boomerang. It's primarily a beaconing framework, so isn't restricted to performance data, but the core plugins that come packaged with it only do performance measurement.

I'll write more in later posts about what you can do with boomerang, but for now, go get it from github.com/yahoo/boomerang. Fork it, use it, patch it, tell us what you think of it.

There's also documentation along with the source code, or you can have a look at my personal copy of them — frequently, but not automatically updated.

I announced it at Velocity 2010, and they made a video of my talk.

While working on boomerang, I needed to get at the current document's URL. The way I generally do that is via the location.href property. This worked perfectly until I hit upon a URL with a # in it. The # was correctly URL encoded to %23, and most browsers returned the correct string in location.href.

Safari 4.x was the only exception. It returned the URL after decoding. This wouldn't be a problem on its own, except that only the URL portion was decoded. location.search (everything after the ?) and location.hash (everything after the #) were untouched.

This meant that a URL like this:

http://tech.bluesmoon.info/foobar%231.html#123

Would show up in location.href as this:

http://tech.bluesmoon.info/foobar#1.html#123

This was a problem, because I now had two # signs in the URL, and would have to twist through hoops to correctly encode the URL. The bug is actually in Webkit, and affects Android too. Thanks to Ryan Grove for pointing me to it.

Then I discovered the document.URL property. This is exactly the same as location.href, except that it isn't affected by the bug in Safari 4.x.

I've tested in other browsers, and it seems to work as expected, so that's what I'm using for now.

Oh, and if you don't know what boomerang is yet, wait until June 24.