Pre-combined sieving remaining.

[Brucifer]

In shooting emails back and forth with Lars, I have learned that besides the 23T left before we are in the combined ranges, there is also 4.7T in first pass ranges. He said he'd like to get the first pass stuff out of the way if possible, so I went ahead and reserved those ranges. As my systems complete the presently loaded 2nd pass stuff, I will load the first pass ranges in to the nextrange.txt files on my systems and start getting the first pass knowcked out. Then I will go back and complete the 2nd pass ranges that I have reserved. I estimate around a week to get the first pass stuff completed.

Just didn't want y'all thinking that I was leaving the 2nd pass 23T remaining stuff. It's all stuff that needs to be finished to get PSP to the point of sieving the combined ranges with SOB. :)

And thank you to those that have reserved some more of the 23T stuff!!!! And if any of you new team members would be interested in spending a month helping us out, I'd be more than happy to help you get set up and running for the psp sieving effort. :)

Bruce

[NeoGen]

I remember seeing those ranges there but didn't know they were really needed for the SOB merging.
I did one once (98000-98700) just to close the hole, and didn't get on them anymore.

[Brucifer]

yeah, I didn't know about them either. But we'll get there, as the ranges left to go are getting smaller each week. :)

[ltd]

To make it clear the first pass ranges are not the most important for the merge with combined sieving but as they are the lowest firstpass sieve ranges they should contain more low n factors then any other range. So these two ranges have a good chance to remove k/n pairs that would else be prp tested in the next few weeks/months. Which brings us ahead in the short term.

For example the combined range 133T-134T done by shauge contained 161 factors lower then 20M and four of them 2.1-2.4Mil. (The llrnet server hands out work in the range 2.05-2.25M at the moment)

For the ranges Brucifer reserved i would expect around 200 low factors per 1T and around 250 Factors>20M.

Conclusion these two ranges are the most important ranges the project has to offer at the moment.

To complete my priority list:
Second priority the 23T from second pass
Third Priority combined ranges.

And yes i do combined ranges most of the time.
I want to show to the SOB sievers that we want to help them and not only want to get their help. I am lucky at the moment that there are so many helping hands to give me the freedom to sieve ranges with lowest priority from my list.

Lars

[AMDave]

I made the switch from PRP to Seive last night.

It looks like I have a total of around 1350 kp/s chewing through my first range. About 20% of a 400 range is done in 24 hours, so I am pleased with the progress.

As I expect to keep topping up my nextrange.txt files for a while, moving range to range to range, I want to be able to send in intermediate results without allowing any of my clients to run out.

Is it safe just to "cut" the fact.txt and factrange.txt files from their directories while the client is still running on each machine, or does the Sobistrator provide another means for doing this like the Rieselator does?

Brucifer, I notice you have been sending in frequent updates, do you cut the files or cut the text from inside the files? Do you only send in the facrange.txt files at the end of a reservation or is the content directly linked to the content of the fact.txt file?

It would be REALLY cool if someone fixed up the Sobistrator to submit PSP Seive residues just like the Reiselator.

[Brucifer]

Say I take out a 1,000 block for sieving.

I seperate them into blocks of fifty. For instance the range is 50000 - 51000, the first block will be 50000 - 50050, the second is 50050 - 50100 and so on. I first start one system off with the first block, then once it is running, I stop it and load in the nextrange.txt so it reads:
pmin=50050000000000
pmax=50100000000000

save it, and then restart the sieving. I use linux, in console mode, and watch then. When that system completes the first range and has shifted into the second block, I stop it, and then copy the fact.txt to 50000-500050_fact.txt and also copy the factrange.txt to 50000-50050_factrange.txt. I edit the files to remove the factors that are out of range. Then I mv the 5* files to a seperate directory and after I get a bunch of them I tar them up and send them to Lars in the format of results.brucifer.20060303.tar.gzip and lars can then use his script to suck them right in. The 20060303 is year month day.

I then re-edit the nextrange.txt file to enter in the next range and start the client again.

Windows people can use sobistrator to take care of the bulk of the babysitting. I don't use sobistrator. Anyway I generally send them in every day or two. When I finish sending in everthing in that reserved range, then I add the endtry to the sive reservation forum and say it is complete.

You do not have to wait until everything is completed to send them in. And it isn't a good idea to do that anyway as that way if the HD or something goes kaput you don't lose a bunch of timetaking sieving results. I also retain a copy of the email message that I send to lars with the results just incase something happens and his system craps out or loses some results too, or also in case I made a mistake and forgot something.

I also save a copy of the range numbered result files on the computer they were crunched on, just in case I screw up and number something wrong somewhere along the line. :) I do not edit any fact* files while the client is running if that was one of your questions.

[AMDave]

Thanks Brucifer.
That makes sense.

I break up my block of 1000 into 1000 :D I keep track of progress of each machine in a workbook and so I can load-balance the work depending on each machines speed (they are all different). I seem to get away with topping up the nextrange.txt file without stopping the clients.

Thank you for explaining how to deal with the in-progress fact.txt files.
Incidentally, I wrote ftp scripts to copy the directories back to my main PC where I back them all up, and I cron'd the job to run each hour. I use the same script when I plug in the laptop when I get home from work. So I have an hourly backup copy of each machine's folder and progress.

I can easily pick up all the fact files from there.

I thought about writing the reverse so that I could modify the files locally and send them back to each machine, but I don't like the idea that I may cause the machine to do a range twice, so I stick with the hands-on method too.

[Brucifer]

ya.... some things are just better left to the hands on approach. Also due to the time it takes to re-crunch a range, I just don't like to add complicating methods to it. :)

I'm getting close to having half of the first pass stuff finished though. Will be glad to get back in the 60000 - 80000 range stuff again though. :)

[Brucifer]

I just turned in some more results. don't know if it is enough to get us past Rechenkraft or not, but it will be pretty close anyway. Hopefully this week we will edge by them. :)

[AMDave]

I think I have my ranges rolling over nicely.
I just dropped in my second batch.
Lars has processed them.
I'll have another batch mid-week.

Waiting for the stats to rollover tonight.

Did you see the last update?
Only ~15 Tera left !!!
We're knocking it down fast now.

Awfully good, Chaps. Tally Ho!

[AMDave]

There we are...only 12 Tera left now.
lots of changes in Proth I see. Lars is busy :lol:
Still waiting for the stats. I'll catch them in the morning.

[Brucifer]

AMDave -- Lars shot me a note and said that between what you turned in and what I turned in that we were only 10,000 points behind Rechenkraft.net on the sieving. We should make that this week and move AMD_Users into our richly deserved number "ONE" spot! Really getting close now after all the hard work!!!

Now if the stats page would just reflect that... :shocked!: It must be that itty bitty bit of Rechenkraft.net shock in him that is slowing down his public acknowledgement of our achievement. :)

\:D/

[Ototero]

Our own Top Guns. Nice going guys.

[Brucifer]

Ahhhhhhhhhhh <very huge cheesy grin> now the stats page reflects it!!!!!!!

Now if we can just squeeze off a few more..

:pottytrain5:

[AMDave]

I see it. Phew. Very close now. :D

I have the "pump" operating well now and have fine tuned my dashboard.
My next batch should be done in 1.8 days.

I have some ground to make up because I arrived so late.
I'll be seeing this through to the end of the ~12 Tera gap.

What happens after that?
Do we crack open the SOB range? :twisted:

[Brucifer]

After the 12 T I will most likely stick around and do some of the combined stuff. After all, just cause we make it to #1 doesn't mean it's there permanently. :) I will also probably put a couple more systems on the Xtremlab stuff again. Said I wouldn't do it after they dumped the stuff from the first run. But it worked first try with no problems which is a lot more than I can say for several of the other boinc efforts out there.

And I think that the PSP effort needs to contribute some of the combined stuff with SOB so that the SOB players get something in return for doing the combined which is really more of a help for PSP than it is SOB at present.

[ltd]

You should have a look at the stats today.
It is really worth it this time. :wav: :greenjumpers: :notworthy: :notworthy: :notworthy:

[NeoGen]

1 AMD Users 48857 5459131.692202
2 Rechenkraft.net - Germany 750998 5337616.159008

Woohoo!! :D :greenjumpers:

[Brucifer]

Finally got there. :drinking:

[vaughan]

Well done.

[Ototero]

I waited and waited before doing the stats, but the page was static :cry:

But it WILL show up in GOLD tomorrow.


Well done Bruce and all those who helped.

:wav:

That's 3 number 1 spots :D

[AMDave]

An outstanding effort Bruce.
Congratulations everyone.

The dashboard is refreshed

We now have 3 Gold - w00t :!:

Thanks Ototero !

Let's mop up that 12 Tera :twisted:

[Brucifer]

Well thank you for all your gracious comments about my efforts. However it truly was a small group thing. Without you others sticking with it, I wouldn't have remained pushing as hard as I could at this. So it *is* a team thing gents. :D

[Empty_5oul]

well done all involved :D

[Lagu]

Hi

Great and well done Brucifer and co!

Lagu :-({|= :thumbleft:

[Brucifer]

We are getting there, but it sure is starting to seem like it's really dragging by the closer we get! :) I just reserved another 3.8T. I see AMDave took out a healthy range too which is really great! :)

It would sure be nice if we could finish this by the end of the month. May be pushing it, but it would sure be nice. I'm anxious to try out some of the compbined stuff and see how that goes.

[NeoGen]

Even when we close the gap there are still a few ranges that are way late, like bwhite's 30900-31000, and Paranoid Androi's 47100-47200.
Those are also necessary for us to totally close the gap, aren't they?

[Brucifer]

I'm sure that as we approach finishing the other ranges that Lars will either sieve those himself, or ask one of the other participants to sieve them. There's not much there really.

[ltd]

These very small gaps make no difference for the real combined effort. But i wrote to both holders of the open gaps last week to see if they are still working on it. For one of the ranges it was finished but the mail either did not arrive or my spamfilter made something wrong for the other range i have no answer so far.

But as said before with only 200G in size these gaps will not stop the combined effort.

Lars

[Brucifer]

Down to 3T now!!! Lookin good!! So we really are getting close to reaching one of Lar's goals! :wav:

[Brucifer]

This is a *really* expensive hobby! :( It's just unfortunate for my wallet that I get such a kick out of it. My recalcitrant A/C unit decided to puke it's guts yesterday afternoon and I had a "scram" session due to rapidly rising high temperatures.

Anyway I was fortunate that the gal at the A/C shop took pity on me and squeezed me in today so I could get my farm running again. So what went wrong? We don't really know for sure but are suspecting a leak. However after crawling along and checking the pressure line, we couldn't find anything. So more coolant was added, and we will find out for sure before long since the hot weather is right around the corner. The last time he was out here working on this the weather was totally terrible, so he isn't sure whether or not he actually put enough back into the system after some pressure switches were added to the system.

So anyway, I guess this ought to cheer Lars up some. :) At least I should be able to get the 3000 block finished. So I'm back in operations with a flat wallet. I keep wanting to buy a dual core, but it seems like somthing keeps cropping up that takes my sandbagged dual core funds..... :(

[Brucifer]

Well it looks like all the pre-combined blocks have been reserved, so now all we have to do is to finish the reserved blocks we have. So it's nice to see that we have reached a PSP goal!!! :)

[Brucifer]

What's the game plan now that thecombined sieving has been reached?

[NeoGen]

They reserved another large batch for combined sieving at they yahoo group, and took out the regular dat file, so I guess the future is in combined. :)

But since combined sieving is quite slower than regular was, if we pressed some more on the PRP'ing side to help find a prime it would be great to all sievers, as each prime found cuts down on the dat file and sieving gets much faster.

[Brucifer]

yeah, I'm averaging about a 25% slow-down on the combined stuff as compared to the psp-only.

Finding a prime cuts down on stuff, but there sure aren't many of them found. And cpu efficiency wise, doing llrnet on an AMD is rather slow, as really it is an Intel game.

[NeoGen]

I think that on more recent AMD's (AMD64's X2's, Opterons) that rule doesn't apply no more... :P

Can anyone here do a few performance tests to see if this theory holds? :)

[Brucifer]

yep, I did, my amd64's are losing 25% too. The combined sieving is deeper and broader than the psp-only sieving.

[NeoGen]

Oops... my question was misleading, sorry. :oops: I meant to ask if the LLR on newer AMD's was still significantly slower than on equivalent Intels.

[Brucifer]

I have only a few pre-combined ranges left that should clear this afternoon on AMD-64's. I will put one of those 64's on the psp llrnet and see how it compares. I know a while back I was doing some rieselsieve llrnet on a 64, and it didn't keep up with my Celeron D 2.8 533. I don't expect that will change on the psp llrnet either. It is just an Intel game unfortunately. Doesn't mean though that you can't use them on llrnet, it's just that they aren't being used most efficiently.

:)

[Brucifer]

I'm running about 3ms per iteraton slower on a socket 939 AMD-64 3200 than the celeron D 2.8 x533. So there is a lot of other stuff I'd rather run the amd-64's on than llrnet.