Poker-AI.org

Re: cfrm cloud computing

2016-08-17T06:54:49+00:00

Thanks for the answers. Anyway my app need of 1gb of RAM but during the excecution it need to save houndreds of gb on the memory.

I took a look at the free version of amazon web server. For now i've decided to do the first version of my bot using sampling to estimate the parameters that i need if it'll have some potential at the table i'll invest more on it.

Statistics: Posted by AlephZero — Wed Aug 17, 2016 6:54 am

Re: cfrm cloud computing

2016-08-13T21:54:54+00:00

When I was working on cfrm, I bought a server off of ebay with 72GB RAM (HP DL380 G5, 2x Xeon L5520) and adapted my abstraction to fit into that memory footprint. But that would set you back about $1k. Made my code in C to optimize memory and performance.
Code and server is collecting dust now, but provided a very profitable ROI.

Statistics: Posted by epo — Sat Aug 13, 2016 9:54 pm

Re: cfrm cloud computing

2016-08-13T10:14:12+00:00

viewtopic.php?f=26&t=2928

Statistics: Posted by spears — Sat Aug 13, 2016 10:14 am

Re: cfrm cloud computing

2016-08-12T19:12:21+00:00

I agree with spears regarding using EC2 instances. You will be better off buying used servers. You can get a dual quad core Xeon with 32 gigs or more for less than $150. If you are insistent on using the cloud I would start with the typical companies: amazon or google. I believe both offer free versions with limited resources or time.

Statistics: Posted by kixs — Fri Aug 12, 2016 7:12 pm

Re: cfrm cloud computing

2016-08-12T16:47:25+00:00

I'm impressed that you get the problem to fit in so little memory. Hardly seems worth worrying about putting a 1GB app in the cloud. The answer really depends a lot on how long you expect your app to run for. I also have reduced the memory to something quite small, but I've come to the conclusion it is cheaper and easier to buy a couple of used games pcs, than to use amazon ec2

Statistics: Posted by spears — Fri Aug 12, 2016 4:47 pm

cfrm cloud computing

2016-08-11T13:19:11+00:00

Hi guys congratulations for the forum and thanks to spears, some of my classes are based on his code.

I'm not an expert of computer science but i've developed a version of cfrm applied to poker in java, that seems work but i can't run it on my limited machine so i need some advices.

I need to use an array were i can put the result of every hand confront given the flop, dimension are 1326x663x52x25 where i need to save three possible states (win, loss, tie). On java with the less expensive primitive type (byte) it costs 1,15gb , in alternative i can use bitfields of 2 bit for 286mb total.

Then i need to use another 450mb for the Paul Senzee perfect hash table to build this table and another 300mb for the objects of my programs. Then my programs is completely a recursive tail call to itself (one for every brunch of the abstract game three) so i can take full advantage of multiprocessors architectures.

If i want rent computing power in cloud whats kind of specifics i need to check? Have you any site that you would recommend?

Statistics: Posted by AlephZero — Thu Aug 11, 2016 1:19 pm