I was at a friend's house last weekend. He's got an eight core (Core2, 3GHz) Mac Pro 2.1 with 8 Gigs of RAM.

Fast, right?

Wrong. Sometimes when he clicks on a folder in his file manager it takes 20 seconds for the system wait cursor to disappear.

Seriously pathetic, right?

He asks me, "Why does it take so long when I've got so much horse power?"

I speculate and mention the kernel article I read a couple of years ago that talks about the microkernel arch and how the user latency is pretty darn high, but after some thinking about it this week I think it has to to more with the mounting and unmounting of his external hard drives than anything else.

While we were talking about it I had this weird brain wave.

Since most modern systems have more than one core and can generally handle multiple threads I wondered about having the user interface speculate on the users next move while waiting for the users next move: a speculative or predictive interface. Its kinda like a branch predictor from processor design.

The idea is: if user double clicks on a folder in konqueror, the system could speculate on what he might do next and carry out the work necessary to do that action, without actually enacting it, before the user actually double clicks. The system could even keep track of common actions and predict based on past history.

The system would determine the outcome of various actions before the user clicks and when the user finally clicks, act immediately on all the data that was generated by the predictive thread.

This would make the GUI seem faster, especially for slow devices or large programs.

I figure someone has already figured this out, but I thought I'd ask the experts.

Is this currently done anywhere in any GUIs? I realize it would eat power like crazy, but I wonder if it would be worth it to make the gui seem less sluggish.

Raydude

I'd offer one of my processors for this...

But the thing that made you start this thread should not be solved this way; 20 seconds to mount a drive on an 8 core sounds like a bug to me...

OhReally wrote:I'd offer one of my processors for this...

But the thing that made you start this thread should not be solved this way; 20 seconds to mount a drive on an 8 core sounds like a bug to me...

I agree its a bug. Has anyone thought of anything like this? Can any persue it? It probably requires an arch change, but it would be really interesting to spawn all the options at once and see how much resources it takes...