Factoring, Sieving, PRPing and ???

[IronBits]

I'm one of those that just run the client. :)

However, I see that there are others doing these 'other' things... and being curious, was wondering if there is a SIMPLE explanation of each, exactly what to do if I wanted to do something other than 'JRTC' (Just run the client) ;)

I have P3s
I have P4s
I have XEONs
I have AMDs
I have AMD64s
I have AMD FX55

Which is better, for the most part, for each method you use.
What are the downsides (like no stats) (why not???) ;)
What are the upsides (we can reach the end quicker)

If I wanted to do something different....
How the heck do I get started, for each method ????

Where's the FAQ - all about SOB but was afraid to ask link? :cheers:
For the 'simple minded folks' like me :rotfl:

[Frodo42]

I think MikeH's All Things Sieving (http://www.aooq73.dsl.pipex.com/) is a good place to start to get an idea about what factoring and sieving is about.

[Mystwalker]

Maybe this thread helps:

http://www.free-dc.org/forum/showthread.php?s=&threadid=7432

Unfortunately, we don't have a decent explanation of factoring.
Just look at it as another way to find factors (like sieving), but you chose a desired test instead of a whole range.


In a nutshell:

To most help the project,

- P4s (and Xeons) should do PRPing (that's the technique of the "main" client) or factoring,
- P3s and Athlon XPs are best at sieving and
- Opterons & Athlon64s are good at both - I don't know where they are of more use.

But remember that sieving and factoring won't give you a prime - these techniques can only erase tests that would have been PRPed otherwise. So every PRPer gets a slightly better chance finding a prime.

[vjs]

Ironbits,

Your probably not going to get a response right away b/c everyone is feverishly updating there clients, checking stats etc... So I'll be a marter, if you want to be a smarter too follow my advice on your machines.

First off for the P4, run prp (JRTC) :D ...

No, really for the P4 JRTC but if since you are dedicated I'd consider being a true marter and logining in as supersecret. You will be doouble checking n-values around 1.3m, the tests are very fast and it increases the lower bounds of the tests range. This slowly increases sieve speed etc. If you'd like to test something a little bigger login as garbage, these tests are around 4.5m take half the time and fill in holes etc.
Neither of them will yeild you any type of personal stats, however you'll get a warm fuzzy feeling knowing you are doing something different than 99.9% of the users.

For the Xeon's thats a tough call, if they are P4 style xeons with HT and small cache 512k or less and limited memory 256 maybe 512mb. I'd say stick those on garbage or supersecret as well. If you have large cache and alot of memory 1G or greater definetly do P1 factoring.

I'll explain these all later....

If the xeon's are the old generation P3 or less stick them with the rest of your lot in sieve.

AMDs, AMD64s, AMD FX55, xeon p3 or less, P2's, P3's, Pentiums, athlons, thunderbirds etc. These should all go into sieving for sure!!!




Here is a copy of what I sent to the ars forum....

------------------


I'm sure alot of people are going to read this thread so I'm going to make a few short beginer statement about seventeenorbust (sob) and the way the project works.

SoB consists of two portions,

First part PRP testing,
This is the portion of the project 90% of the people run. One sets up the client and tests a particular K/N pair for primality. Once that one test is done it sends in its results. The majority of the numbers tested are not prime. Our 8th prime is expected around, 10.1M-16.8M but these numbers are just a guess, we could get another in a week.

"Primes often come in groups"

Second portion of the project deals more with eliminating k/n pairs from testing (prp).
If any k/n pair has a factor it cannot be prime and therefore does not need to be tested by prp.
Eliminating k/n pairs through factors is currently much faster for the project as a whole since every k/n pair will be tested twice.

K/N pairs can be eliminated by either P-1 factoring or sieveing.

Sieving removes alot more k/n pairs than P-1. However alot of the factors removed by sieveing won't be test for years to come, but they are removed. Factoring removes less k/n however it removes those that will be tested within a few weeks.

In either of these two cases factors that are easily verified are produced eliminating the k/n pair entirely, no (slow) prp testing required.

As an individual the second portion of the project will never find a prime, but will help the first portion find primes faster.



Also there are some hidden features to seventeenorbust.

If you login as either garbage, or supersecret, you'll be able to tests some very fast small numbers. However these are teamless accounts and you can't get personal credit.

supersecret tests numbers around 1.3m tests take <10% of the time of a regular test 7.8m.

garbage tests numbers around 4.5m tests take <half the time of regular tests.

What these accounts do is, double check the lowest k/n pair, supersecret, or test the lowest untested pair, garbage.

It is possible that either of these account will produce a prime. You will probably recieve credit for the discovery, but it's won't help our team stats.

Thanks any questions about either of these topics please ask in the team threads.

------------------------

In any case the P1 and sieve do have stats they are not as predominate as the prp stats but they do exist.

http://www.aooq73.dsl.pipex.com/scores.htm

As for setting up the clients isn't it alot more fun to figure things out on your own???:D Seriously if you look around its documented.

It will probably take me longer to write this than you, searching and setting things up on your own. Here goes.

For sieve...

Go to the reservation thread in the sieve forum it's a sticky.

1. Create a folder somewhere on your computer call it sieve proth cmov whatever...

2. Download the dat file and unzip it into the created folder creates a file called SoB.dat

3. Download/extract the zipped client called proth cmov.exe , rename it if you wish into the same folder.

4. Reserve a range to sieve, something like 601000-602000 Ironbits [reserved]

5. Click on the cmov icon within the folder it will ask you start: type 601000 then press enter it will then ask end : type 602000 press enter. The client will start and your sieving.

6. for time to time go to that folder and open fact.txt submit the contents through a copy paste into the following window. http://www.seventeenorbust.com/sieve , it's also recommended to do the same for factrange.txt, factexcl.txt is totally upto you I submit it as well.

For P1 it's pretty much the same thing,

1. Go to P1 factor thread

2. create folder, download, reserve,

3. Heres where it's a little tougher, using a dos box, navigate to the folder and type the following.

sbfactor 802000 803000 49 1.5 756

This basically tell the computer to factor all k/n pairs in the range from 802000 to 803000 (the one you reserved) the number 49 and 1.5 are sort of weird... They represent how the client should test, just go with these numbers until you understand why. The 756 is the amount of memory to dedicate to the program if you only have 256mb total in your machine I'd say don't P1. If you have 2G then enter something like 1500, always leave about 256MB for your machine. 512mb could use 350 without problem, 1G what I have I specify 756.

Hope this helps

Ask questions if you need help

[Keroberts1]

thats not a matter of the client being slower. The units cem/sec are not a measure of computing power. For smaller N values the cems will be lower but the rate is the same. if you tests numbers in the 31337 range you're cems per sec will be much higher. its just a attribute of the types of numbers we're testing and the method we're using.

[engracio]

I concur with Stromkarl's finding. Did the same thing on this box and it came to a grinding halt. Will run it for a while but will go back to main prping later on once it completes it current k/n.

Don't know how it runs under PRP Client v2.2.0


e:(

http://www.clanhosts.com/dev/sobsig/sig.php?engracio


Never mind Keroberts1 answered the question while I was pecking on my keyboard.
:blush:

[Keroberts1]

try experimenting with a 31337 account for a little I wouldn't reccommend putting much effort into the test though cause finishing a test would take about 3 times as long as a normal prp around 8000000 and the tests might be eliminated before the prp bounds ever gets to it.

[royanee]

cem/s are a flawed unit because they do not accurately represent the difficulty/computation ability based on the data used. Just as Keroberts said, with low n, the number is lower, and high n, the number is higher. This is due to the difficulty increase being approximated originally as n^2 (or something similar) when it is actually nlogn (or something similar). If the growth rate was fixed, then you would see the same number on all tests. I believe this was a planned improvement to the v3 client.

[Keroberts1]

look at the number of blocks ina test per size of N value

[IronBits]

Thanks to everyone who participated in this thread who actually helped to make it more clear for nubies like me in this project, and stayed on topic.
To the rest of you, please take your arguments to another thread because it's just confusing the issue.
Please stay focused to the thread topic. Don't post :|ot|: ...
Hopefully there is still more to learn about the different things one can do to assist in the project. :D

[Mystwalker]

Originally posted by Stromkarl
That cannot be! This machine has been running at 150-180k cEM/s for over a year! The cEM/s do not change, just the total number of blocks to finish a test.

Well, actually, the cEM/s increased a bit due to growing N, but it most likely stayed within the margin of recognition, as the increase in N was low (maybe 2M) compared to the current distance to supersecret tests (about 6.5M).
I'm not sure, but I think the bias is of a relative and not an absolute nature (sloppy speaking), so we have an N increase of 1.35x vs. a difference of 6.24x - this would have quite an impact, which can be seen after all.

So don't worry about this behavior, your computer and the PRP computations are ok, only the measure is not suited. ;)


Also, as a result of my experiments, SOB staff should expire two supersecret tests:
n21181 k1258172
n55459 k1258186

supersecret tests have an expiration time of 1 day (hey, they complete within a few hours!) - so there's no need to expire them manually. :)

[vjs]

Ironbits,

I'll go with you on the off topic post but thread like thse come and go pretty quickly.

let me just sum up all that was said above, sure everyone want's quick results of how fast and maybe I wasn't clear.

I was speaking with respect to total time...

Run one test using normal login you'll get something like 3 days
Run one test with garbage probably get something like 1 tests per day
Run one test with supersecret and it will finish in <2 hours

Ignore cem/s totally what's important is how long anyways, It might be useful to test how long verison 1.2 vers 2.3 take at n=1.2... But they are probably the same.



Back on topic,

Ironbits, what have you decided to do with your machines, have you figured out sieving... having a fresh look from a newbie is always great!!!

I realise it's not self-explanitory but was it simple once your overcame the activation energy and the first few hurtles?

Let me know I'd like to know, there is also a benchmark thread somewhere in the sieve client thread it might be nice to post there with your speeds.

[Stromkarl]

Ironbits and vjs,
Actually, my original post was on topic. I asked why I was having trouble using one of the three ways to help SOB per vjs' reply to you. I had some trouble with it and asked for help. If it will make the situation any better, since your original topic is very pertinent to my situation, I have deleted any of my posts that may be off topic.

I have (4) P4s, (6) PIIIs, (1) PII and have noticed now that the PIIIs and PII are not helping SOB much with the PRP effort. After reading vjs' suggestions, I attempted to try the most painless method of helping out - switching to supersecret or garbage. I was shocked to see that there was such a large drop in cEM/s. After I looked at my SOB logs and read your reply to my problem, I understand it much better now. I will probably switch (5) PIII-550Mhz, (1) PII-400Mhz over to supersecret soon.

Sieving seems to be very good way to help with the computers I have, however, it also seems to require a lot of sneakernetting to deal with submitting all the text files online. I don't mind that at home - so I will setup (1) PIII-1Ghz up for this. Unless I can figure out a better way to submit all the data, I do not believe I can afford to take the time with the computers at work.

Thank you for posting very pertinent topic, and your help!

Stromkarl

[vjs]

Stromkarl,

Perhaps my comment about "off-topic" were not totally justified... your post was a good one and a good question, but IMHO it deserved it's own thread.

This project is very specific and often times people will be talking about one thing and then the topic goes of on a slight tangent (cem/s). The thread then answers a question not realted to the post subjet.

In anycase I might be interesting if you create a new thread regarding cems and post your finding new and old client etc. You'll probably get better answers if you need them, but I think you figured it out already correct.

Glad to see that your going to run some supersecret with those P3's, as for sieving your correct again... Look into a programs call sobistrator I think it will takecare of all your needs if the computers are networked.

Let us know,
VJS