Buffering work

[N.V.M.]

any chance this project will ever be able to cache work? i'm asking because some of my team mates are on dial up and don't have a constant connection.

[em99010pepe]

I have the same problem and I use this software:

http://www.adslkeepalive.co.uk/

Hope that helps,

Carlos

[N.V.M.]

thanks, i'll pass that on but would like to get an official answer to the future possibilties of caching work from the creators of the project.

[Bok]

From some mails I've seen from the project owners it's just not really feasible due to the type of work being done. When a new unit is requested it is generated at that time based on the current point of the simulation.

You can't really create units ahead of time due to this constraint.

Bok

[omer]

yea Bok is right, the server changes its configuration after some workunits and then any previously assigned work is useless, so cant buffer!

[alpha]

omer, what are your (and/or the other guys) views on slower computers participating in the project. I have a P2 233 on eOn and it takes around 45 minutes to complete a job.

[omer]

The slow machine will get you points for every submitted result, but I doubt the results will be useful for the server (I will have to check with Graeme to see how often the state of the server is changed), if its less than 45 minutes, then basically you are not contributing anything, the results are ignored.

[alpha]

omer, it would be great to have that information. I have a variety of slow computers crunching eOn at the moment, but if they are not contributing useful work I will happily put them somewhere that they will.

It might be worth putting something in the FAQ on the website, such as a minimum CPU speed as a requirement. Not that it would make much difference to the stats hoes. :)

[graeme]

A 233 MHz machine is pushing it. As people have mentioned, the server works on a task for about 30 min to an hour. After this time, the server uses the information from the clients to move to a new state. At this point, any old work is not useful to the project. I set this time to accommodate the slower machines, but only to a point. If more time is spent on each transition, the calculation is more accurate, but less overall time is simulated. This is also the reason caching will not work for the project. A fast machine can complete many work units per transition, so in principle it could cache 5-10 WU, but any communication gain is outweighed by the chance that the server will transition to a new state, and all the work units will be lost. The network load for clients on the project is not high, but the nature of the calculation only makes sense for machines with a permanent network connection.

[alpha]

Thanks for clearing that up graeme. I'll pull the plug on the P2 233.