Linux versus Windows ratings

[rsbriggs]

Well, unboxed some old dual PIII-600 systems (just for fun). First machine, Windows XP and a FaD install. After a full day, each processor running under Windows XP is rated at a speedy 44. Hasn't budged from that.

On the second machine, I popped in a copy of Cluster Knoppix, (the only recent live CD I have) downloaded FaD for Linux, and FADProgress. Fired it up. One day later, both of those nodes are rated a blistering 51. Haven't moved from that mark at all.

I'll grant you the fact that these are not exactly blazing FaD crunching machines - but the point is - notice that the linux nodes are rated almost 20% higher.... I get 14 more points per hour, or an extra 336 points per day out of the one with Linux on it :D

[LAURENU2]

Also linux cuts though a job 5% faster then M$ so that would up your increass of points to 25% to 30%
But as I said linux missed hits windows would pick up

[PY 222]

Originally posted by LAURENU2
Also linux cuts though a job 5% faster then M$ so that would up your increass of points to 25% to 30%
But as I said linux missed hits windows would pick up

LAUREN, the missed hits on Linux, will it affect the results that we are sending back to the server?

Does THINK know about this issue and if so, what is he doing about it?

Sorry for shooting out so many questions at once but this missed hits on Linux thingy is bugging me as my pharm is 90% Linux and I sure don't want to be doing crappy science.

[LAURENU2]

. LAUREN, the missed hits on Linux, will it affect the results that we are sending back to the server? No not really returns from both OS are OK


. Does THINK know about this issue and if so, what is he doing about it? Yes he knows and nothing will be done about it he just works with it


. Sorry for shooting out so many questions at once but this missed hits on Linux thingy is bugging me as my pharm is 90% Linux and I sure don't want to be doing crappy science. The difference in hits are vary little like 1% or less
Vary minor
I believe think said that the extra hit M$ gets is because the OS has more time ti sift through the Mol

But I think The difference in hits also changes with O/C to high The higher the O/C the more hits are missed

[PY 222]

Thanks for the info Lauren.

But this "missed hit" thingy still bugs me as what if one of those "missed hits" were actually something useful.

Do the FaD server dish out the same job to at least 2 different machines or more for validation purposes? Kinda like what BOINC does with their workunit?

[PCZ]

The current bias in granted credit in favour of nix is obviously a bone of contention and needs looking at.

Think :Pokes:

However
Regardless of the OS used the work we are doing is good valid Science and the slight variations between the OS's are inconsequential.

Arguing about this is non productive.

Consider how much would be lost to the project if PY222 decided not to run think because he listened to some old wives tale about linux not finding as many hits as windows and concluded his contribution was flawed.

Who's to say that the supposed extra hits that the windows boxes find are all valid.
Think doesn't consider this to be an issue so let that be our guide.

It would be a good thing if Think could adjust the CPU ratings to bring windows and nix more in line.
This disparity is contentious and could easily be avoided with a tweak to the CPU rating code.

[FluffyChicken]

The missed hits are due (most probably) to timeouts.
THINK times out if it takes to long in a calculation against molecule, the time out is based on CPU Rating (so slow computer should do as many calculation as fast computer before it times out (a CPU Rating of 100 would timeout after 60mins, 200 after 30 mins etc..)

So as most Linux (not all) systems still seem to be getting a higher rating than a comparative Windows rating there timeout will be sooner and this may mean it misses that hit. BUT that hit wouldn't be important anyway.. hence why he's not bothered.

(I remember something about then next client (1.30) may have an improved system to reduce mollies getting into the timeout , or that may have been the rotations :looney: )

I'm not sure that Linux systems go through a job faster (I always thought they where slightly slower as the calculation efficiency wasn't as good for FaD, , that's assuming no timeouts which would speed the job completion up)


It'll be hard to get the ratings similar and is one of the problems of cross platform computer which gets even worse in the Linux world as the 'core' setup vary even between distro, bsd, deb, kernal revs, compilation, blah blah.


Just think of it as encouragement not to use Microsoft OS's ;)



Lauren2, CPU Rating is Points/hour of crunching, so you'll still get the same points in one hour of crunching no matter how fast it cuts through a job (for the same rating).
So you cannot add the % together.