Comment by pron | Hacker News Reader

Comment by pron | Hacker News Reader

Parent Full thread

Some public benchmarks are old, but they were done after Dalvik had gotten its JIT (both runtimes have seen huge improvement since), and after it had been "optimized for mobile": https://blogs.oracle.com/javaseembedded/entry/how_does_andro...

You are right that Dalvik was written to be optimized for mobile, but you have to understand compared to what. Dalvik is optimized for mobile compared to simple VMs that don't have a lot of work put into them -- not compared to HotSpot. Java's embedded versions have had a lot more effort put into them than Dalvik.

Dalvik itself was plan B for Google, after they'd failed to secure a licensing deal with Sun (and didn't want OpenJDK's GPL). It wasn't anyone's first choice, and certainly wasn't some new powerful technology. Just a VM that was good enough compared to the effort that's been put into it.

There are open source (obviously) OpenJDK distributions[1] specifically made for embedded devices (although you can build them yourself), and (proprietary) editions of Java have been running on devices as constrained as SIM cards for much longer than Android exists.

[1]: http://www.azulsystems.com/products/zulu-embedded

You are glossing over the fact that Dalvik is a multiprocess VM. There is no such JVM, even today.

This aspect and the Dalvik bytecode are the main reasons why Dalvik is so fast on mobile. No other VM (or JVM) can realistically compete, especially not HotSpot. Not now and certainly not back then.

> You are glossing over the fact that Dalvik is a multiprocess VM.

Making HotSpot multi-process would have been a much smaller effort for Google. It really isn't that hard.

> No other VM (or JVM) can realistically compete, especially not HotSpot. Not now and certainly not back then.

I've just shown you how HotSpot embedded handily beats Dalvik. Remember, Dalvik was Google's plan B. It wasn't anyone's first choice, and certainly not some cool groundbreaking technology. It's a not-too-shabby VM for the amount of effort put into it (which isn't much), and it was available for Google to buy. That's all.

Why would you want a multi-process VM anyway?

Was the is the performance difference between HotSpot and Dalvik worth $45 million in 2006 dollars?[1] Those were the alleged licensing terms proposed by Sun ($20 million upfront + 10% Android revenue capped at $25 million).

Your budget determines the kind of software you get. On a purely engineering basis, Hotspot might have been the better option, Business-wise, Dalvik made sense to Google.

1. http://www.pcworld.com/article/253666/a_timeline_of_oracles_...

I don't know, but it was certainly worth much more than the $0 they would have paid for HotSpot if they'd had the guts to go with the GPL. That's what's missing from the timeline:

* March 2007 -- Sun open sources the JDK at the perfect moment for Google to adopt it. Google chooses not to because they don't like the GPL.

I think the phone vendors would have caved. It might have taken another year, but they would have realized they had no choice once the app store opened in July 2008, as none of them could have had the brand recognition to build an attractive software platform to rival the iPhone.

I strongly suspect Android's uptake by manufacturers and MNOs would have been far less if it was licensed as GPL. So there would have been an opportunity cost to Google (that is certainly not $0).

Do you think Sony, let alone Verizon would cave and agree to the terms of the GPL? Remember this is in 2008 - Sony's DRM/rootkit scandal was only 3 years prior (2005).

Edited: fixed my years

Verizon doesn't make phones (I think; not too familiar with them), so what do they care? But Sony, yes, I think that when the app store opened in mid 2008, they would have realized they have no other choice but to rally around a single software platform, and one that's made by a company with a bigger brand in the software world than them. If GPL was the bitter pill they had to swallow, they would have. The upside to GPL from their perspective would have been that all of their competitors have to abide by it, too.

Also, just to clarify, the pill really isn't all too bitter. Use of OpenJDK wouldn't have dictated that all apps must be GPL -- not even bundled apps made by the manufacturer. The OpenJDK has a clear "classpath exception", which makes it more like LGPL -- any code dynamically linked with it is not required to be GPL. The only implication is that any modification they make to the actual runtime itself would have needed to be GPLed. I don't think phone manufacturers make any technologically groundbreaking, game-changing changes to the runtime -- anything that could hurt their competitiveness had it been free.

So I can understand Google's decision, and they certainly believed going GPL was an opportunity cost, but in retrospect, choosing OpenJDK would have -- at the very worst -- delayed Android's adoption by a single year, but would have ended making everyone's life much easier. I think that in hindsight, it's clear that passing on OpenJDK was a big mistake. Don't fear the GPL.

Verizon does not make phones - it sells them. It was responsible for much of Android's early success in the US with it's multimillion dollar "Droid does" ad campaign to the extent that "Droid" was synonymous with "Android".

You really haven't seen embedded Java have you?

http://www.atego.com/products/atego-perc/

Just one example.

Reply on news.ycombinator.com