One day to go are we going to do it?

[Fozzie]

The question is, will Fozstradamus be wrong:shocked: :scared: :shocked:

[PY 222]

Its gonna be close.

Don't know for sure until tomorrow.

/me crosses fingers

[CaptainMooseInc]

Well, it's gonna be close.....as of 2:19PM EST we're sitting less than 2 mil gflops behind BeOS and 3mil gflops behind US-D?

Anyone have a clue as to how many GFlops/Day we're going atm?

-Jeff

[PCZ]

What is the target time exactly
Do we have to catch US'd before the 25th or on the 25th ?

PS
We could carry on till the new year and roast that chicken :jester:

[Bok]

They are trying, but it's futile :) I think it will be late tomorrow, perhaps I should switch my remaining 30Ghz or so later today...but if it's Christmas Day then I think mission accomplished. We should push on somewhat through the end of the year and probably get up to perhaps 11th ??

Stats should be online fairly soonish I hope.....

Bok :D

[em99010pepe]

I will help you guys. Just switched my two OGR machines to FAD.
It's not too much but it's what I can do for now.

Carlos

[Fozzie]

mission accomplished as far as Fozstradamus says.

Awesome team spirit we have here.

:notworthy Free DC

From the official site we are 352,714 points adrift and overtaking at 593,250 a day. So midday tomorrow should be overtake time.

:cheers:

[ronbo54]

Well so far in almost 2 days I have 1 result showing. Sounds a bit slow for a 3GHz HT box. It either froze or I'll get a huge number of points all at once. :confused:

[Fozzie]

and that'll tell you whats going down and a predicted end time and points/GFLOPS return.

[jasong]

This is probably going to cause some wtf posts, but the secret to knowing the approximate(usually within 5-10%) score you've earned so far is dependent on the fact that at least until 2 weeks ago, and probably still...

The rating that FaD gives is based on hourly points instead of GigaFlops. I forget which selection it is, but if you have the window that shows the rating and the second count(this count will probably go slightly slower than clock time), you can calculate your score:

seconds/450*rating

This equation comes from consolidating the equation:

(seconds/3600)*rating*8

GigaFlops is almost certainly a PR scheme, but they didn't bother(that I know of) to change the rating method in Windows. :bang:

[PCZ]

Or more simply

A cpu rating of 100 = 2400 points a day.
200 = 4800 points a day.

This is assuming that the client has the CPU all to itself.


CPU rating x hours = points.
points x 8 = gigaflops

[em99010pepe]

I'm crunching two huge units! Is that normal?

The estimated points for each one is more than 40,000 points.

Carlos

[Fozzie]

they do exist.

I've had 70k off one WU before

[em99010pepe]

Originally posted by Fozzie
they do exist.

I've had 70k off one WU before

I'm asking because today, 24/12, I don't think I will finish those units.


My units:

Project: Cancer
Hits: 82 and 156
Time (s): 101,000 and 72,000
Progress: 49% and 63%


Carlos

[Fozzie]

and US-D have ramped again.

Do they not realise we are on a schedule here.

This could be very tight.

EDIT

The figures hadn't come through totally at that time they have now and we have slam dunked US-D.

Grand job Free DC FaDDers

[alpha]

Originally posted by Fozzie
they do exist.

I've had 70k off one WU before

70,000 points, or gigaflops? Must've been one hell of a job if it was 70,000 points!

Currently working on a job which has been running for 150,000 seconds. Using PCZ's formulae above, it should be good for ~60000 gigaflops (~7500 points) at the moment. Going strong at 90.73%.

[Fozzie]

Can't remember the query, I think it was an HIV one and did nearly 2 million seconds on a barton 2.6 GHZ, over 1000 hits too.

[alpha]

Nice one Fozzie :thumbs:

The job I mentioned above finished after 166,075 seconds, scoring 8396 points with 256 hits. Not bad. :)

[Fozzie]

That should pop you up a rank:thumbs:

[alpha]

Heh, not even in the top 10 daily within the team!

Back when I first started FaD, you specified the speed of your processor, and appropriately sized jobs were handed out. How does this work in the current version? The reason I am asking is because my LAN gateway is a P2 233 but I don't really want it to crunch FaD if the jobs are going to take weeks+. Any hints?

Also do we have any idea if there is any performance difference between Windows/*nix in the current version?

Starting to give this some more thought, because not being in the top ten daily within the team will just not do, eh? :)

[FluffyChicken]

the cpu speed did not distributed quick/slow jobs at all. All it did was determine a queue size. This was abandened at v1.23 clients (I think) in preference to the newer send when completed option (you just use the connect every xx hrs/days now). It was changed to ease load on fads servers.

course most people want the queue size control back :)

[PCZ]

alpha

I am not sure which OS actually crunches faster but you will score more points in linux.

On the same hardware linux rates the CPU about 15% higher than windows.

I believe Think is trying to even up the scoring between the OS's, they made an adjustment in 1.25g to lower the linux CPU scores but it is still too high.

PS
That job you mentioned above that scored over 8000 points, was the average CPU rating somewhere close to 182 ?

[alpha]

Originally posted by FluffyChicken
the cpu speed did not distributed quick/slow jobs at all. All it did was determine a queue size. This was abandened at v1.23 clients (I think) in preference to the newer send when completed option (you just use the connect every xx hrs/days now). It was changed to ease load on fads servers.

course most people want the queue size control back :)

Oh? My mistake.


Originally posted by PCZ
alpha

I am not sure which OS actually crunches faster but you will score more points in linux.

On the same hardware linux rates the CPU about 15% higher than windows.

I believe Think is trying to even up the scoring between the OS's, they made an adjustment in 1.25g to lower the linux CPU scores but it is still too high.

PS
That job you mentioned above that scored over 8000 points, was the average CPU rating somewhere close to 182 ?

Interesting. Kind of unfair that there is a bias involved.

Indeed, the job I mentioned was running on an XP 2200 at ~1.9GHz with an average CPU rating around 180.

I have a Celeron 466 running FreeBSD. I may throw the Linux client on it and see how it does. However, something is still unclear to me. If some jobs take hundreds of thousands of seconds on modern hardware, how long are these jobs going to take on sub-GHz hardware?

[PCZ]

alpha

A 466 Celeron has a CPU rating of about 30.
Six times slower than your 2200 XP.

So a days work on the 2200 XP is a weeks work for the Celeron 466.

[LAURENU2]

Originally posted by alpha
Oh? My mistake.



Interesting. Kind of unfair that there is a bias involved.

Indeed, the job I mentioned was running on an XP 2200 at ~1.9GHz with an average CPU rating around 180.

I have a Celeron 466 running FreeBSD. I may throw the Linux client on it and see how it does. However, something is still unclear to me. If some jobs take hundreds of thousands of seconds on modern hardware, how long are these jobs going to take on sub-GHz hardware?
I have a 1 old printserver here 200 MX a job takes about 35 days wo compleat:shocked:

[alpha]

Originally posted by PCZ
alpha

A 466 Celeron has a CPU rating of about 30.
Six times slower than your 2200 XP.

So a days work on the 2200 XP is a weeks work for the Celeron 466.

Sorry, I didn't make my question clear. I understand how the stats work, I just don't understand why the jobs aren't sized appropriately for differently performing hardware.

Perhaps this project just isn't suitable for old hardware? I just don't remember having these thoughts when I had a P3 450 on the project a while back. The jobs didn't seem to take long at all.

Maybe I should just shutup and get on with it and see what happens :)

LAURENU2: :thumbs:

[wirthi]

Well, alpha, you get more points when jobs take longer. Just like at OGR (distributed.net); when a paket is twice as big (takes twice as long), you get twice the points.

What I don't like about FAD is that CPU-Rating thing. Don't know if that really works the way it's supposed to.

[FluffyChicken]

It works as well as any other project using a performance rating system, it's just easier to compare or seems to be.

There's no other fair method to use than the 'work done' method.

Time for jobs have changed as client refinments alteration are done but some of the latest HIV & Cancer jobs have taken forever even of high rating CPU's.

The problem they have is they don't really know how long a job will take until it's been done, they can estimate from a small sample the average job length (as they do) and it is in the requested list from many people :D ;)

But in general don't do the HIV/Cancer Jobs on the slower computers. (eg <1GHz)

[wirthi]

Well, yes. With distributed.net that's easy: so many keys, so many stats units. So many nodes, so many stats units.

Of course it's much harder with this project. You can't measure hits - sometimes you work for days on one query and get a few lousy hits, sometimes they pop out every minute. Molecules are not that significant neither.

Well, it's no use anyway, my P4s will keep their bad rating :)

[alpha]

Originally posted by wirthi
Well, yes. With distributed.net that's easy: so many keys, so many stats units. So many nodes, so many stats units.

With OGR in mind, the range of job size is not so great.

You might get a 300Gnode stub or a 1Gnode stub. Even on fairly slow hardware, the largest jobs take days, not weeks. If you are crunching a stub and it gets to 230Gnodes, you can almost think "well it probably won't be much larger/longer".

I've never been able to think like this with FaD. Not that it really matters, though.

[FluffyChicken]

Originally posted by wirthi
... Well, it's no use anyway, my P4s will keep their bad rating :)


If the P4 could somehow get a better FPU* for the MHz then they would get a better rating ;) and get more work done


*more specifically (afiak) the single point fp's

[LAURENU2]

It can't be right WoO stats have us listed to overtake XPC in 38.3 days ?
http://woo.gotdns.com/FrameStats/mainframe.php?Team=2091&Show=Orig :)

[black_civic55]

if we can get a few more people to help us it'll be quicker then that!!

[Bok]

Originally posted by LAURENU2
It can't be right WoO stats have us listed to overtake XPC in 38.3 days ?
http://woo.gotdns.com/FrameStats/mainframe.php?Team=2091&Show=Orig :)

I guess they have our average WAY up based on your 73M+ coming in the other day :)

It'll settle out in a few days on my stats and (as long as our average is higher than their's) should show a legitimate number..

Bok