PDA

View Full Version : Anyone else?



Darkness Productions
03-26-2004, 11:58 AM
Anyone else with a high-speed P4 want to jump into the mix here? I have some very basic stats up at http://stats.dp.cx/gimps/ to keep some track of people... I notice that there's me and I think one or two others running it right now.

We even got into the top 200 of the project!

rshepard
03-26-2004, 12:09 PM
I've still got one box working this-- had it off for a bit while I was seeing how much
I could produce in DF running "flat out", but I resumed prime hunting on it yesterday.

MrTRS
03-26-2004, 12:57 PM
DP,
I have a P4 2.66 ghz with a gig of ram that I was planning
to put on it when the fast DF came up.

I think I'm not going to wait for the slower DF to come back.

I will put it on GIMPS today.

Ray

IronBits
03-26-2004, 01:55 PM
I have a P4 2.8 GHz laptop on it.

Darkness Productions
03-26-2004, 02:34 PM
Wootage! Maybe this will help push us a bit.

wirthi
03-26-2004, 02:49 PM
What makes a P4 fast on this project? I have a plain and simple P4 1.5 with - I guess - 533 FSB and 512 MB SD-Ram. What counts here? CPU power or FSB/Memory speed?

I've run GIMPS before, perhaps I'll return. How long would it take to complete a standard workunit ("Lucas Lehmer Test") on my machine?

wirthi
03-26-2004, 03:03 PM
Ok, guess you convinced me, I'm on that again. PC is running the obligatory one-hour-test now to prove my PC's not calculating shit ...

But I'm a bit confused by the "End User Information" Screen (in the Menu Test->User Information).

I have 5 Textboxes there. That's what I've entered:

Your name: Christian Wirth
Your email address: wirthi@gmx.at
Your user ID: FreeDC
Your password: go-4-it
Your computer ID: wirthi_p4

Is this what I'm supposed to enter to be both on the team and able to view dp's stats with my one name on them?

Darkness Productions
03-26-2004, 04:02 PM
That should be all you need. The stats update just after Midnight EST every night...

ECL
03-26-2004, 06:07 PM
DP, where are you getting the numbers for P90 CPU years on the completed exponents? I haven't been able to find this info in the various primenet files, which has pretty much dampened my enthusiasm for building my own stats engine for this project.

I'm hoping to see P90 totals by user soon, if only to see how badly Nineball is trampling me.

IronBits
03-27-2004, 12:18 PM
197th place!!! :cheers:

wirthi
03-27-2004, 12:27 PM
How long does a workunit take? I've completed "stage 1" on my first, then came "stage 1 GCD" and right now I'm through 30% of "stage 2". It's been long since I last ran gimps: IIRC there are three stages with the third being the slowest by far?

Darkness Productions
03-27-2004, 01:05 PM
wirthi - You may want to switch to running double-checks. They run a little bit faster, and count in the overall stats.

ECL - I found a calculation for an approximation of the P90 years (what's displayed isn't exactly right) somewhere on mersenneforum.org...

Condor
03-29-2004, 09:35 AM
http://www.mersenne.org/status.htm

To calc p90 years, use the link above.

Locate your exponent in table under range column. Go across to find PII-400 Speed.
Multiply that exponent x 5.5 x PII-400 column then divide by 31536000 which is the number of seconds in one year.

So wirthi's P4 should net 3.2 years for the exponent 23693977.

23693977 x 5.5 x .776 / 31536000 = 3.206

Fozzie
03-31-2004, 01:20 PM
and its answered a few of my questions.

How many stages are there?

rshepard
03-31-2004, 01:24 PM
IIRC, there are 3 stages if you are doing LL-testing. there are 2 "pre-factoring" stages and then the third stage is the actual LL-test if no factor is found.

guru
03-31-2004, 02:05 PM
Ok, I've got 9 P4 3 gig systems running Linux that I want to run this on. They are all on a private network with no internet connection. Can someone give me a some info on setting up these systems and what is needed to transfer the work to and from the systems.

I have been running RC5 on these systems and using a USB drive to transfer the work to a internet connected system. I want to do something similar with gimps but I'm not sure what is needed.

guru :confused:

rshepard
03-31-2004, 02:15 PM
See if this (http://www.mersenneforum.org/showthread.php?s=&postid=15155#post15155) helps.

heretic
03-31-2004, 02:15 PM
How to do sneakernetting

To sneakernet (nonet) or move work in progress around various machines, follow this procedure: The prime95 client structure is quite simple. There are 5 basic files that are in use:


prime.ini
local.ini
worktodo.ini
the checkpoint files p123456, q123456 and pl123456. the first two checkpoint LL tests (one is a backup, to nonet only one is needed) and the last one checkpoints factoring work
prime.spl, which may or may not exist


Optionally, copy results.txt and prime.log if you want to see what was going on while the work crunched. The safest way to handle sneakernetting is to install each instance in a unique directory and configure it as such with its own userid. Start up the client, get the amount of work required (keeping in mind you must make an update within 60 days maximum) and then stop it. Zip the above files to move them to the new target install.

On the target box, install the client as torture test only. This ensures that there will be no confusion, as all the info required is found in local.ini and prime.ini. Alternately, the entire client can be copied over from the source box.

Prime.spl may or may not be present depending on the work type and settings. It contains the results (and unsent messages to the server). If it exists, do not let it and the checkpoint files become unsynchronized. To dump a box, stop the client and keep it stopped until you return with fresh work. Then, it will pick up exactly where it was last stopped on the exponent in progress. When returning to the nonet box you must delete the old prime.spl (and checkpoint files if you had run the exponent for any period of time on the master box, otherwise do not copy back the checkpoint file as it will not be needed).

For nonet boxen it is desired to minimize the number of communication attempts. Set the send new end dates in options / preferences to 58 days and set the network retry to 300 minutes. Set the cache option to 1 day to prevent the client from attempting to get more work.

heretic
03-31-2004, 02:23 PM
Along the same line:

Can I move to a different computer and continue work there?

Use the following simple steps:
[list=1]
Stop the program on the old computer
Copy the *entire* file folder to the new computer
On the new computer, add any startup shortcuts you want
Start the program, set the CPU info in the Options settings
Let the program contact the PrimeNet server to send an update
[/list=1]

...and

Can I switch work units between two machines?

Yes you can. Just follow the steps below:
[list=1]
Stop both clients.
Back-up both clients directories.
Swap both client directories.
Restart both clients.
Re-config both clients with correct CPU and memory details.
Clients automatically contact PrimeNet with new completion dates
[/list=1]
Easy, clean and works like a charm.

heretic

Fozzie
03-31-2004, 02:58 PM
Installed it as a service on one P4 2.8 and will have it on another of the same power.

Once I have checked how it all runs I'll push out to 3 more.

See how it goes from there.:cheers:

willy1
03-31-2004, 03:24 PM
I'm in for a few P4 2.8 Xeons

DF has driven me nutz lately anyway :)

heretic
03-31-2004, 03:31 PM
With the power FDC has you guys might still give tpr a good run for the money :D

heretic

guru
03-31-2004, 05:05 PM
Thanks, I got the 9 systems up and running. They are still running rc5 but once they finish all the keys from my key server then they will be dedicated to gimps. I also have my desktop P4 1.5 running gimps.

Finally a project that takes advantage of the P4's. :cheers:

guru

Fozzie
04-01-2004, 06:45 AM
famous last words :D

My first ox says it has finished a LL test? Thought that was a bit quick?

Anyways I will try and upload tomorrow and see how it goes.

Gonna set up another 2 clients today probably.

rshepard
04-01-2004, 07:29 AM
If it actually ran the LL test, then yeah, that's a bit fast :D
If it found a factor in stage 1 or 2 though, then it would skip the LL portion, and that is normal.