New Client is very fast

[PCZ]

It's raining points.

Now to RK anonymous :D

[em99010pepe]

I've been saying that for the last two weeks. :bang:

Just arrived to the TOP100 using the windows client!

Carlos

[IronBits]

Originally posted by PCZ
It's raining points.

Now to RK anonymous :D :rotfl: :smoking:

[PCZ]

The linux client still has the advantage speed wise.

My not very scientific test says linux is twice as fast.
About 3 mins a test for windows and 1.5 mins for linux on an athlon xp2400.

[em99010pepe]

Originally posted by PCZ
The linux client still has the advantage speed wise.

My not very scientific test says linux is twice as fast.
About 3 mins a test for windows and 1.5 mins for linux on an athlon xp2400.

If I am not wrong the wu's aren't the same size.
The important here is that now I can run the windows version without losing too many ground to the linux users.

Carlos

[magnav0x]

Someone should try running the linux client in Cygwin on a windows system and see how well it will run.

[IronBits]

It runs on Mac OS X to!

Someone with a Mac Mini run it and let us know your results!

[IronBits]

Originally posted by magnav0x
Someone should try running the linux client in Cygwin on a windows system and see how well it will run. $ ./client
./client: 1: Syntax error: "(" unexpected

[magnav0x]

Just tryed it myself as well IB, same error :(

[PCZ]

"Cygwin is not a way to run native linux apps on Windows. You have to rebuild your application from source if you want to get it running on Windows. "

The above rule isn't set in stone is it ?

[IronBits]

:bonk: :scratch: :umm: :o

[magnav0x]

Well I've run plenty of linux native apps on it. I suppose it's their "excuse" for not all apps being compatable.

[PCZ]

Thats what I thought.

[Longbow]

Originally posted by PCZ
The linux client still has the advantage speed wise.

My not very scientific test says linux is twice as fast.
About 3 mins a test for windows and 1.5 mins for linux on an athlon xp2400.

Being the type that never likes to be at a disadvantage I fired up a Knoppix live CD and tested out the Linux version of Eon for a few hours and haven't seen any noticeable difference.

:dunno:

[PCZ]

I did say my test was not very scientific.

Maybe someone can do testing on identical hardware with both OS's.
Anyway the windows client is much better than the old one ;)

[graeme]

It's true that the linux client is much faster than windows. The heart of the calculation is done in fortran, and there are not the options for windows fortran compilers that there are under linux. Still, at some point we will have a faster windows client. There is no intrinsic reason that it should be slower. The new windows screen saver, which does not use opengl, also takes up less cpu time which will help the client speed.

Some aspects of this project make timing tests a little difficult. One is that the individual work units do not take the same amount of time. There is quite a bit of randomness from work unit to work unit, but also over time. This is because each client is searching for a different process available to the system. They are given a search direction, in a high dimensional space, and they wander until they find a saddle point. Sometimes this is quick, and sometimes it takes a while. Another reason the work unit time is variable is that it depends on the system size, or how many atoms are being simulated. Right now, it's a pretty small system, and the work units finish quickly.

[PCZ]

Thanks for explaining.

[FoBoT]

Originally posted by graeme
Right now, it's a pretty small system, and the work units finish quickly.

so in terms of strategy , this is analagous to a small protein in DF, a good time to put the pedal to the metal on EON to gain ground on the competition. if a "large system" comes under study, then we can aford to move assets to other projects, if desired

[graeme]

Certainly, from the point of view of getting the most work units on any project, this makes sense. However, you could also focus on a project you are interested in and take the fluctuations as they come. One thing that separates this project from lot of other DC projects is that it's not a trivial calculation that can be split up evenly into millions or billions of independent work units. Instead, everyone is working on the same calculation at the same time. The work that a client does affects the state of the calculation, and everyone's work thereafter. If someone finds a favorable process, for example, the system will move in that direction, and everyone will start looking for ways to move from that new state. It is much closer to the concept of having a large distributed super computer to make a calculation run faster, instead of doing many independent calculations towards a common goal.