How Do Browsers Scale?

An anonymous reader writes "Benchmarking browsers is a somewhat silly exercise, since scores cannot be replicated on a variety of hardware, and it is not uncommon for even the same system to fail to replicate benchmarks scores, especially in JavaScript tests in two succeeding runs. The guys over at ConceivablyTech have an interesting approach, running browsers through multiple tests on different sets of hardware (including an Android smartphone), and showing the scaling differences between browsers when you are using a dual-core netbook on the low-end and a six-core desktop on the high-end. They also tested HTML5 on Firefox mobile and found the browser has better HTML5 support than the current Firefox 4 Beta 6."

Comments

Need a better client-side scripting language

[Ynot_82]

Why, oh why did javascript become the defacto client-side scripting language for the browser
If you want to scale horizontally across multi-cores, you need a language that allows easy multi-threading and concurrency
About the only thing JS offers for concurrency is that horrid settimeout function

What we need is a better scripting language
Why not incorporate a Python interpreter into browsers, and develop a stripped down sub-set of python for use on the web
I see no technical issues in doing this, only trying to battle the inertia of JS

Re:Need a better client-side scripting language

[rantomaniac]

Wait, what?

Python as a choice of multithreading-enabled language? You are aware that CPython has a global lock and only one thread can execute Python code at once?
Javascript will be more multicore friendly than Python when web workers get widely implemented.

And what's the point of developing a brand new sub-set of python with a brand new interpreter and set of libraries? You might as well just compile python to javascript, there's not a lot of impedance mismatch between them. Python is mostly useful because of its wealth of libraries, other than that it's just a generic dynamically-typed language with a certain syntax.

Re:Need a better client-side scripting language

[spinkham]

Javascript runtimes are way faster then most other scripting languages at the moment.

If you go to the benchmark game you'll see v8 is about 6.5x faster then python and tracemonkey is 3.8x faster.

The only credible "scripting" language runtime that is faster then Javascript is LuaJIT, and Lua has nowhere near the features that JavaScript does.

Re:Need a better client-side scripting language

[kripkenstein]

Why not a VM that runs bytecode, and JIT compiling, so your favourite language can be used?

Several reasons. First, all such VMs - the JVM, .NET, Mono - have far worse startup performance than JavaScript. That's why Java applets never took off. While those VMs improved in that respect, they still lag far behind JS in startup time. They are optimized for long-running processes, not small client scripts that should get going instantaneously. Different goals here.

Second, there is no clear open standard for such a VM anyhow. The JVM and CLR are known to have various patents around them; we have recently seen Oracle sue Google, and Microsoft execs have admitted that Mono is only free from risk because it is licensed by Microsoft. So neither is a good foundation for something meant to be truly open and free, like the Web should be.

Third, existing JavaScript engines are, by far, the fastest implementations of dynamic languages out there (much faster than such languages running on the JVM, .NET or Mono - Jython, IronPython, etc.). IMHO it makes sense to have a low bar for writing scripts for websites. You shouldn't need to do C/C++ memory management, or write lots of static typing definitions, to add a little scripting to your website. So something like JavaScript, Python or Ruby makes sense. But again, this last point is clearly debatable, that's just my opinion.

Re:Conclusion: Firefox 3.6 scales best across core

[icebike]

Nope.

The only takeaway you need is:

Chrome 8 had the smallest gain, which, however is due to coding flaw in the Sunspider benchmark that holds back the processing horsepower of the Phenom II X6 processor in general.

Translation: Our results are totally bogus because our tool was broken but rather than fix that, we are just going to shovel these results out there anyway.

dual-core netbook on the low-end?

[nurb432]

That's NOT low end. Funny how marketing is so skilled in manipulating peoples perceptions.

Re:Dual core netbook is low-end?

[c6gunner]

I was surprised by that, too. Dual-Core is my high end, with a PIII on the low end.

Cool. So what's the weather like in Uganda, this time of year?

Yay bench

[kangsterizer]

Another meaningless benchmark that claims to replace all the previous meaningless benchmarks. Yawn.