Like Chad, I think MagLev’s initial performance numbers will hold up. It’s possible that as it matures it will get slower, but it could get a lot slower than it is now and still easily be the fastest Ruby VM around.
And I agree with Chad that it might even get faster. I’ve also spent some time investigating how to make Ruby run on a Smalltalk VM, and it’s a really close fit. During the presentation at RailsConf, either Avi or the Gemstone guys revealed that they had modified their Smalltalk VM by adding two new bytecodes aimed at Ruby. I’ll go so far as to speculate: it’s likely that those two bytecodes deal with variadic methods and creation/lookup of dynamic instance variables. It sounds as though the core Ruby language is nearly complete on top of that base, so it’s easy to imagine that the early, hurried implementation of those two new bytecodes could be optimized further. And some of the Ruby features that have hurt JRuby’s performance will be no problem on a Smalltalk VM—ObjectSpace, for example, can work using the same facilities that Smalltalk’s development tools use today.
The persistence story is amazing. Avi and the team at Gemstone plan to implement an interface that is similar to ActiveRecord, but cleaner, since the object/relational impedance mismatch no longer applies.
Finally, there’s the question of licensing. I’ll be shocked if MagLev is open-source, but I think there’s room for a proprietary Ruby implementation. The team has committed to complying with RubySpec, which means I’m not very worried about compatibility.
Most Ruby projects won’t need MagLev, but the ones that do will gladly pay for a top-notch, supercharged implementation with great scalability and persistence stories.
I’m definitely looking forward to hearing more about MagLev over the next few months.