20051117, 14:52  #1 
Feb 2003
Istanbul
34_{16} Posts 
distributed project search
hi
I'm looking for a distributed project for my 5 P4 boxes (3 of them do not have connection). Now they are working for Riesel Sieve Project but because of some reasons it's no more a fun project (for me). I've been following PSP for more than 3 months and PSP is one of the projects I want to join. But before joining I want to confirm some points below 1. Can the admins guarantee they won't abandon my ranges without a warning? (I lost more than 50 k/n pairs in Riesel Sieve so that's not fun.) 2. Are the stats reliable and updated regularly? Will I get credit for all my works done? 3. Will the admins share their plans with the contributors? (I don't want to be forced to double check like in SoB. That may be I good idea but I want to have some choices.) thanks for the answers drakkar67 
20051117, 15:52  #2  
Apr 2003
2^{2}·193 Posts 
Quote:
To point 1: The handling of reservations depends on the way you make the reservation. For sieving your reservation will stay until you have finished your range. The only thing i expect is that i receive a mail every month or so saying " i am still working at that range". If i don't hear from a contributor for around 2 months i will try to contact him via forum or if possible mail to get an information if he still wants to finish the range. For PRP testing there are two ways possible. a. You make your reservations via the llrnet server. The server is configured to distribute a workunit automaticaly to somebody else if it is not returned for a week. b. You make a reservation in the PRP reservation forum for a range of your own. Then the same rules as for sieving apply. So nobody will reshedule your range until you tell me that you have abandoned your range. To 2: I try to update the stats every day. But to be honest due to the fact that this is still semi automatic there can be times when i can not manage a daily update due to real live problems.( This is normally a maximun outage of one day) When there is a stats outage due to a holiday on my side i drop a note in the forum ( See another forum entry about this weekends outage) For points given for your work the following rules apply. For sieving: The first person returning a factor gets the point. So if by chance two siever on different reserved ranges find a factor for the same k/n pair the first returned result gets the points. ( This is something that is really rare) If you return results for a range reserved by somebody else your results will be dropped. For PRP. If you are on the llrnet server: The person with the reservation gets the points. So if you return your results after the seven days the next person gets the points. For manual PRP reservation: The person who has the reservation on the forum is the only one who can get points for the tests. If you return results for a range reserved by somebody else your result will be countwd as second pass tests. To 3: I will do my best to make all decissions in cooperation with all the contributors. I made once the error to think that a in my point of view neccessary stats modification was clear to everybody and then i found out that some of the users stopped contributing due to this modifications. So i think learned from this. Double check on this project for example will allways be a manual thing. So you can decide if you want to do it or not. There will be no saying that anyone must do PRP,p1,ecm, sieving.... You will allways have the choice. I think for you a manual reservation of some ranges would be the best as you have no automatic timeout. Disadvantage of this is that you will have to make some things manual. I hope my answers help you to decide if you liek our project enough to contribute. If you have more questions please ask, Lars P.S. I will be out of town from friday to sunday so answers either today or sunday evening. 

20051117, 16:17  #3 
Jun 2003
1,579 Posts 
Just to mention that if you wanted to run LLRnet but wanted to return results in a few months then, you could also run your own LLRnet server.
Citrix edit: if you do not mind my asking, what is the speed of your p4's? Last fiddled with by Citrix on 20051117 at 16:21 
20051117, 22:20  #4  
Feb 2003
Istanbul
2^{2}×13 Posts 
Quote:
I don't know exactly but I think 1 P4 2.8 db core, 2 P4 2.66 (laptop), 1 P4 1.6, 2 P4 2.4, 1 P4 1.6M (laptop), and some more P4's but I think its hard to use them in dc's because I don't want to force my employees. And I don't exactly know how effective I can use those machines. I have some retired PIII's. If I can find a place I'll put them in sieving. And 2 G5's, 2 G4's and 1 G3 (which can be used at weekends) but sadly I couldn't find a good math project for them. I can say that these machines can test about 2530 k/n pairs of 196 K FFT length daily. I've contributed GIMPS for about 7 years with just one machine. But that's an infinite project. SOB, riesel and PSP are (theoretically) finite projects. So I joined SOB but I hated its client and the way the admins' choices. I joined Riesel but I lost about 50 pairs because of some faults. I join these projects JUST FOR FUN. But loosing 50 pairs (because of whatever the reason) is just FRUSTRATION. So, before making a decision I wanted to clarify some points. Actually, I don't want to contribute just by cycles but also for example I want to run a Turkish (I live in Turkey) version of a PSP web site or something else. I have one more question. Maybe I can find the answer in the forum but I have no time to search for it. Scoring formula... Does it give the same score for a unit time of work. A bad example is SOB... lower the FFT size, lower the score for a unit of time. thanks drakkar67 

20051118, 04:47  #5 
Jun 2003
1,579 Posts 
You can try the free trial version for mac's
http://www.iemulator.com I do not know if it works or not. Try proth_sieve and PRP. For floating point they say you can get 3.4Ghz etc, see the FAQ. Let me know what happens. Good luck and welcome to the project, Citrix PS: ltd will answer the rest of your questions, I do not have an answer for the stats question. Also you can make a turkish website if you want, I have no problem with that. 
20051118, 04:59  #6 
Jun 2003
1579_{10} Posts 

20051118, 07:38  #7  
Apr 2003
2^{2}×193 Posts 
Quote:
There are two different formulas. 1. Sieving: Points given take into account the n value removed and the factor found. You get more points for higher n and for larger factors. The maxmimum points you can get per factor is 6000. This was done due to the fact that else a lucky "p1" factor would give you far to many points. There is a possibility that the maximum points will be changed in the future. This comes out of two reasons. The 6000 was set when our maximum n was 20m. The second reason will be a new siever that allows higher sieving then ~1000T. 2. PRP: Points are calculated as depending on the n value tested On the long run my tests showed that the formula is fair in the range n>200K to n<15M. For n outside this range the formula tends not to give enough points. For the n<200K this is no problem as most of them are done by Citrix or myself. For n>15M we have several years before we have to revisit the formula to make it fair again. I said the formula is fair on the long run because it does not take into account the points where a fft change takes place. So short before and after a fft change you get a little bit to many or not enough points per n. But due to the fact that there are not that many fft changes and i also have no idea how to put that into a formula that is fair i made the decission to ignore the problem. To make reservations easier for people with higher performance it is also possible to reserve ranges with a special size. On request i can build for eaxmple reservations that span "delta n"=25000 where the normal reservation has a "delta n"=1000 or you can reserve a special k that you like to test from the first unreserved n to whatever n max you want. Hope my answers help, Lars Last fiddled with by ltd on 20051118 at 07:39 

20051118, 11:05  #8  
Jul 2005
2×193 Posts 
Quote:
Siever will work for Riesel, PSP, SoB and 321, although the initial versions will not be able to sieve both Sierpinski (SoB, PSP) and Riesel (Riesel, 321) at the same time but I am planning to do this. I intend to finish off the G4 version in the next few weeks (it will run on a G5 but will be nowhere near as fast the the 64bit G5 version will be). It will be safe to run this sieve to p = 2^51 and, with a bit more work on the Eratosthenes sieve code it should go all the way to 2^63. 

20051118, 11:14  #9 
Jun 2003
1,579 Posts 
http://hpc.sourceforge.net/index2.php
Might help you in writting the sieve. You can make other things than CPU do your work! Citrix 
20051118, 12:47  #10 
Jul 2005
2·193 Posts 
Thanks for the link but I've looked into Altivec stuff before and because of it's single precision limitation it isn't that much help.
To multiply two 64bit numbers you just have too many intermediate values to add up and carry (the altivec would have to perform 36 multiples and then add these up into 6 result registers) for it to be worth while. 
20051118, 13:45  #11  
"Mark"
Apr 2003
Between here and the
3^{3}·233 Posts 
Quote:


Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
AlphaZero style distributed computing project for Go and Chess  MooMoo2  Lounge  5  20190528 13:22 
A distributed ECM project for Fermat numbers  ET_  FermatSearch  1  20170222 10:42 
A distributed ECM project for Fermat numbers  ET_  Factoring  4  20170220 13:22 
Distributed NagellLjunggren Search  TeknoHog  Open Projects  24  20151111 00:55 
POLL  How many other distributed project are you currently  delta_t  Lounge  9  20021210 09:48 