On Dec 7, 2011, at 8:33 PM, Darren Cook <[email protected]> wrote:

> I think you got a burst there; 1 ECU is very roughly 1 Ghz of an Xeon.
> 
> So, the micro instance would be too unreliable for a computer go
> competition.

I must admit I have not paid much attention to consistency of performance as 
it's not our focus at work. As long as it scales. The Libego benchmark only 
runs for a few seconds, so it may not reflect what is required for real Go 
playing. I did get the same results consistently though, also for my Java 
program, which runs a little longer.

> 
> The small instance costs $0.085/hour ($0.12/hour on windows). So if each
> game takes 50 minutes (leaving 10 minutes for setup), and you play 10
> games in your tournament, it will cost someone $0.85 ($1.20 for
> Windows-only programs) to enter the tournament.
> Not unreasonable; the cool thing is everyone could set up 10 small
> instances and all rounds could be played simultaneously :-)
> 
> Downsides are that that is not much CPU, so level of play is going to be

Amazon needs to make money providing the service of course. Still, their tools 
to automatically launch instances and manipulate them remotely could have some 
interesting uses.

By the way, I decided to try to run Libego on my MacPro using VirtualBox and 
got a similar result. Was I wrong in remembering it would do 40 kkps per Ghz or 
thereabouts? Or was that a completely different version?

Mark Boon
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to