Rogers and BitTorrent: another datapoint

[Yanni Chiu]

Lennart Sorensen wrote:
> Virtual machines are too inefficient and mainly make sense when you
> don't know what the archtecture is going to be.

Don't agree. Check out the literature around the
recently open-sourced StrongtalkVM. With dynamic
type-inference, calls are faster than C++ implementations
that use vtables. Once the VM figures out what is
being called, from where, it's a goto statement; whereas
the statically compiled C++ still has to go through
it's vtable. There is proof that VMs can be very efficient;
just because there are lots of slow VMs doesn't mean the
VM approach is inefficient.

 > Most programs you know
> are going to run on a specific architecture, so you should simply
> compile it for that and not some stupid, bloated, inefficient virtual
> machine.

See the "id" work at http://piumarta.com/pepsi/.
IIUC, this work was just presented at OOSPLA'06,
and the technology is a minimal machine to build
virtual machines. In addition to a Squeak virtual
machine, he's done a Javascript implementation
in 400 lines.

--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists