Macintosh sieving client

[Death]

need one.

havin g5 1.8 dual around///

[royanee]

nbeGone (is that the name?) is the Mac one I think, there should be a link on Mike's sieving site. Though, if you directed this at mklasson, you might be able to ask him really nicely for one that runs on Darwin (MacOSX) since it is *BSD based. Though actually, it usually seems like getting the Linux copy of programs to run on *BSD systems is fairly easy... or so I'm told.

[Death]

i got nbe_gon but it havent any readme, so I can't run it. i don't even know is it suitable for sob sieving?

phil don't answer on e-mail yet. hope to hear from it...

[vjs]

IF you ever do get sieve to work death I have two g3 doing nothing.

Let me know.

Don't know much about the g3's nor do I care too... sorry mac guys ... can you install osx and run a linux client, is there a step by step somewhere.

Now if I had a g5 things would be different.

Also what kind of speed would one expect from either.

[Mystwalker]

Unfortunately, I can't remember how to use NBeGon - I used it when the performance race NBeGon <--> SobSieve was running and NBeGon in the lead...
AFAIK, there's someone who used NBeGon on an Alpha system. Maybe the forum search can dig him up and you can ask him for advise.

Anyways, the performance will be very low compared to proth_sieve - NBeGon should be at maybe 50% of the speed and it doesn't use AltiVec. I'd suggest trying to trade work - e.g. by looking for projects where Macs are very good (like distributed.net) and asking someone (ideally from the own team) to trade maybe 100 RC5-72 workunits to sieving a 10G range - if the other person uses a P3/Athlon/... for RC5, I'm quite sure both of you will have an advantage.

(I recently traded some ecm curves of a Mersenne number against PRP tests of PSP numbers - it was a "win-win-situation"...)

[royanee]

You could always try loading up a ppc version of your favorite Linux distro... or maybe you couldn't... actually... if you load a ppc linux, run wine, and then run the windows sieve client, you'll probably be doing fine. Maybe?

[Death]

but it works F%^%N slow!!!

got got a message from fat phil

> well, I'm a SoB siever and want to use a G5 on it.
> i download your client and new SoB.dat file.
> well, I don't completely understand commandline parameters for
> Nbe_Gon_osx_10

Oooh, that was nearly two years ago now - I wonder if _I_ can remember!

> using proth_sieve on PC i enter next parameters - sieve from 123000 to
> 123500 http://www.aooq73.dsl.pipex.com/ - there's All Things Sieveing
> page about SoB sieve.
>
> so what range i must put in a -p parameter in a terms of sob sieve if
> I want to sieve 123000-123500 range?
>
> 123000 - this is in a G's I suppose.

I assume so, so -p=123000000000000-123500000000000

You'd might double-check that G=1000000000 rather than 2^30 though.

> please, help.
>
> and having dually G5 how instances should i use?

Use two, but give them different p ranges.
One with -p=123000000000000-123250000000000, and one with -p=123250000000000-123500000000000.


Phil

=====
When inserting a CD, hold down shift to stop the AutoRun feature In the Device Manager, disable the SbcpHid device.
http://www.cs.princeton.edu/~jhalderm/cd3/

g5 dually 1.8MHz
it have macosx 10.3.5

so i run two instances from different folders like this ./Nbe_Gon_osx_ppc -p=426300000000000-426400000000000 -s=SoB.dat
second with 426400-426500

and it goes.
it start at ~19:00 and now at 10:00 it on a 426001250000000
please calculate its/s somebody for me. too now.

knowing too small bout linux commands i can't run it at background process. it should be something like "nice -19 nohup" but don't know for sure. =))

so g5 goes back to ogr-25 where it screams!

i don't know can i made make and make install on a osx? is there any sources available. but i know very little about programming.

[Death]

Originally posted by vjs
IF you ever do get sieve to work death I have two g3 doing nothing.

Let me know.

Don't know much about the g3's nor do I care too... sorry mac guys ... can you install osx and run a linux client, is there a step by step somewhere.

Now if I had a g5 things would be different.

Also what kind of speed would one expect from either.

you have mac osx on g3's? you can install osx on them if you have a usb on this macs.

what can you do with g5? ogr-25 rule! got ~#25 on a daily stats..

[Mystwalker]

Originally posted by Death
but it works F%^%N slow!!!

second with 426400-426500

and it goes.
it start at ~19:00 and now at 10:00 it on a 426001250000000
please calculate its/s somebody for me. too now.

Seems to be 23 kp/s - back when public sieving started, this was quite fast...

knowing too small bout linux commands i can't run it at background process. it should be something like "nice -19 nohup" but don't know for sure. =))

Something like that (you can leave out the "nice -19" part when NBeGon already runs at low priority; I'm not sure if it does, though...) - I prefer using "screen" - just enter "screen" (maybe you have to press away some text now), and then do what you like. Before disconnect, just press CTRL + 'a' and afterwards 'd' (having unpressed the CTRL key). Now, the screen is 'detached' - it's still there, you can reaccess it via "screen -r".

so g5 goes back to ogr-25 where it screams!

You really should do that. If you want to help SoB, I'd suggest you indeed look for someone using P3/Athlon PCs and do a trade. Look at the ratio in ogr-25 (maybe 10:1) and in sieving (maybe 1:5, too). Then, you decide on an exchange ratio, possibly 7.5:1 - which means that you increase your sieving performance to 750%! AND you increase the ogr-25 performance of the other guy as well!

[Death]

well, but it works with larger numbers than proth_sieve

right now it ends 9999999-10000001 range and it run fast. doing 2G less than in a day.
((2 000 000 000 / 24) / 60) / 60 = 23 148.1481 well, not so fast =((

but I won't trade. I'll run ogr-25 for myself. =)) havin #12 in a OGR-24, two places left for TOP10 and suddenly it ends =(
now my target is TOP100 in ogr-25.

[vjs]

Death are you saying that your only getting 23 kp/s with a g5 in sieve???

[vjs]

Death also are you saying that you can sieve upto (10000000G) this value any any values below...

Hummm...

have a look at this

1702062.04G - 1714286.04G, size:12224.01G, est fact:3121
1714286.04G - 1733181.96G, size:18895.92G, est fact:5090

Within this "small" range the factors are 10 times more dense that our current range of ~500T.

For example my
460000G - 480000G, size: 20000 produced 689 factors
Even if you only get a sieve speed of 23 kp/s it would be equivalent to ~250 kp/s for the factor return. Also the points you'd score for these would be pretty high.

I'd really like to try this range and see if the factors are really that dense.

OR if someone knows of a client for windoze that will sieve 1.7p let me know.

[Death]

Originally posted by vjs
Death also are you saying that you can sieve upto (10000000G) this value any any values below...

Hummm...

have a look at this

1702062.04G - 1714286.04G, size:12224.01G, est fact:3121
1714286.04G - 1733181.96G, size:18895.92G, est fact:5090

Within this "small" range the factors are 10 times more dense that our current range of ~500T.

For example my
460000G - 480000G, size: 20000 produced 689 factors
Even if you only get a sieve speed of 23 kp/s it would be equivalent to ~250 kp/s for the factor return. Also the points you'd score for these would be pretty high.

I'd really like to try this range and see if the factors are really that dense.

OR if someone knows of a client for windoze that will sieve 1.7p let me know.

yeah 25k or so. pity
and Phil give me a sources, but i can't compile them. know too little about mac programming....

but yes. it can sieve up to 54 bit.

well i try some of this but it takes a whole day to sieve 2G so it takes 1000 days to sieve 20000G. there should be one factor on a 4G, so once a two days there'll be a factor.....

well worth to try....

[vjs]

Death,

I tried downloading nBegon from the site having a tough time with the switches etc.

Do you know if this client work on win2000 and the switches required for the 1.7p range???

is it

nbegone -s=sob.dat -p=1700000-1700001

etc

=2^54 should take us to 18,014T correct

[Death]

*G*

you should use G

*G*

-p=1234567G-12341234G

like this...


and I start 1714286 range

and little correction maximum 52 bit not 54

[vjs]

Death not tryiong to do your range etc but this is what I'm getting...

nbegon -s=sob.dat -p=1714286G-1714287G

# DLOG[11/1.35681] Huge (5072*3735)
# 2.8 p=1714286000000023 (#f=o) has overflows 0|0|0|0|0|0

is this correct no speed reported????

Also are the factors simply reported in sob.del ????
no fact excluded etc...

You seem to be informed here this is totally new to me.

[Death]

Originally posted by vjs
Death not tryiong to do your range etc but this is what I'm getting...

nbegon -s=sob.dat -p=1714286G-1714287G

# DLOG[11/1.35681] Huge (5072*3735)
# 2.8 p=1714286000000023 (#f=o) has overflows 0|0|0|0|0|0

is this correct no speed reported????

Also are the factors simply reported in sob.del ????
one fact excluded etc...

You seem to be informed here this is totally new to me.

yeah kinda....

5080*3740
and same numbers counts...

you can't believe your eyes? =))

i can measure speed visually. like a 25kit/s. and doing 2G/day makes same speed.
yep. it also creates a .bat file, for windows users to continue work after it stops.
inside

 ./Nbe_Gon_10_osx -s=SoB.dat -f=SoB.del -p=1714286090000000

useful for paste into terminal window

at least it saves his state sometimes...

and yes there's no factexcl.txt and factrange.txt only factors itself....

[vjs]

O.K. Thanks alot death I started this range

1702062.04G - 1714286.04G

from -p=1702062G

on my fastest machine. I'll let it run for 24hrs and report back on my speed, normally I'm getting ~700 kp/s with cmov.

[Mystwalker]

Originally posted by vjs
# 2.8 p=1714286000000023 (#f=o) has overflows 0|0|0|0|0|0

The numbers at the end show the number of times where the first "try" failed and a new calculation has to done, where the more retries on one number(?), the more right value is incremented.
So 10|5|2|1|0|0 means that 10 times, 1 retry was necessary, 5 times 2 retries and so on.
Unfortunately, I don't remember exactly what is retried...

It is possible to influence this by a different "alpha value" (I've forgotten how to set it for NBeGon, though). I guess some REALLY old threads hold all the information needed...

[Death]

sorry, missed one parameter in a .bat file "-d=1.36" i think that it is.

and there's always this wonderful key -h =)))

so it runs forward until monday.

huh huh

mystwalker.de/sb/fatphil/bin/NbeGon_009_FB
2,1) Lmodqq0,24) buf:r(0,23) j:r(0,1) start:r(2,2) origx0,3) origstart2,2)
count0,1) x:r(0,3) a0,20) k:r(2,5) i00,3) primegen_init.c primegen_init:F

that's first from two pages when search in google for primegen_init.c


sofar i can't make own client. i wonger what speed i can gain just by recompile it with gcc3.3 for 64 bit architecture. Fat Phil say performance will raise significally.

b30570:/primegen-0.97 mac$ make
nroff -man primes.1 > primes.0
nroff -man primespeed.1 > primespeed.0
nroff -man primegaps.1 > primegaps.0
nroff -man primegen.3 > primegen.0
( cat warn-auto.sh; \
echo CC=\'`head -1 conf-cc`\'; \
echo LD=\'`head -1 conf-ld`\' \
) > auto-ccld.sh
cat auto-ccld.sh make-load.sh > make-load
chmod 755 make-load
cat auto-ccld.sh find-systype.sh > find-systype
chmod 755 find-systype
./find-systype > systype
( cat warn-auto.sh; ./make-load "`cat systype`" ) > load
chmod 755 load
cat auto-ccld.sh make-compile.sh > make-compile
chmod 755 make-compile
( cat warn-auto.sh; ./make-compile "`cat systype`" ) > \
compile
chmod 755 compile
sed s/conf-words/`head -1 conf-words`/ \
< primegen.h.sed > primegen.h
( ( ./compile tryulong32.c && ./load tryulong32 && \
./tryulong32 ) >/dev/null 2>&1 \
&& cat uint32.h2 || cat uint32.h1 ) > uint32.h
rm -f tryulong32.o tryulong32
( ( ./compile tryulong64.c && ./load tryulong64 && \
./tryulong64 ) >/dev/null 2>&1 \
&& cat uint64.h1 || cat uint64.h2 ) > uint64.h
rm -f tryulong64.o tryulong64
./compile primes.c
primes.c: In function `main':
primes.c:42: warning: return type of `main' is not `int'
./compile fs64.c
cat auto-ccld.sh make-makelib.sh > make-makelib
chmod 755 make-makelib
( cat warn-auto.sh; ./make-makelib "`cat systype`" ) > \
makelib
chmod 755 makelib
( ( ./compile trylong64.c && ./load trylong64 && \
./trylong64 ) >/dev/null 2>&1 \
&& cat int64.h1 || cat int64.h2 ) > int64.h
rm -f trylong64.o trylong64
./compile primegen.c
./compile primegen_init.c
./compile primegen_next.c
./compile primegen_skip.c
./makelib primegen.a primegen.o primegen_init.o \
primegen_next.o primegen_skip.o
./load primes fs64.o primegen.a
ld: can't locate file for: -lcrt1.o
make: *** [primes] Error 1

does anybody know what all this st*ff means???

[vjs]

I didn't start it with -d=1.36 but it did put it in the .bat file.

Should I restart my range????

[Keroberts1]

i don't believe that the expected afctors for such a hgh range should be that high. That is probably just an error in the reporting of the factor density. i don't remember waht the formula was that was used to determine what ratio the factor density has to sieve depth but it was constantly decreasing i know.

[pixl97]

Death

./load primes fs64.o primegen.a
ld: can't locate file for: -lcrt1.o
make: *** [primes] Error 1

does anybody know what all this st*ff means???

Lets see...

First you want to do
$locate crt1.o

You should get

/usr/lib/crt1.o
/usr/lib64/crt1.o

which should be...

$rpm -qf /usr/lib64/crt1.o
glibc-devel-2.3.3-27

You may be missing your glibc-devel, since im guessing this is linux on a ppc arch of some type it should be as easy as installing that, then again ppc may work totally differently.

[Death]

vjs, this is "self-tuning" parameter, don't worry about it.

Keroberts1, well, time shows. this is more fun than factors hunting. But such a lagre factor will be nice...

pixl97 oh man! there's no such file at hdd =(

locate crt1.o does nothing.
and i have usr/lib but there's no such file. and also no lib64...

Index of /fedorappc/FC2/Fedora/RPMS
glibc-devel-2.3.3-27.ppc64.rpm 11-May-2004 13:35 2.4M

i found this. you think crt1.o must be in this file?
and how I suppose to unpack this file, there's no rpm command in macintosh bash =((

it doesn't work =(((

i have gcc 3.3 20030304 Apple

[vjs]

Hey Death,

Here is what I found after ~18 hrs of testing,

Within the range of 1,702,062G-1,702,072

1702070438636209 | 21181*2^8838572+1

One factor? I'm not sure if it's a unique factor or an excludeded etc... If it's a unique factor I was going pretty good.

My rough calculations on speed yeilds 155 kp/s down from ~700 kp/s, so if the factors are really 5X as dense that would mean a relative ~755 kp/s speed.

Humm, it's a toss up right now. I think what I'm going to do is move back to regular sieving until I can get an answer from Mike or mklasson or I finish my other ranges. If I can get the MAC's up and running I will put them in this range in the mean time.

My final line was

# 68000.4 p=1702072888328201 (f#=1) hash overflows : 811149|39|0|0|0|0

[Death]

vjs why don't you wait for next factor?

btw, progress so far not good not bad

client still don't want to 'make' but Fat Phil give me some advices to make client working.
i send compile errors back to him///
now his turn =))
nice that he sends me sources and help me modify them///

THANKS Fat Phil


btw can anybody mail me crt1.o file?

being away from mac for weekends left it sieving my ranges.....

[vjs]

Death,

I'll be back on this range next week don't worry one day it's enough testing,

I'd just like to finsh up some of my other ranges first and potentially get an answer from Mike. Also submit the factor and see if it's unique.

It's possible nbegone is just testing those factors in the dat file and doesn't produce outofrange or excluded. If this is the case then one factor in 18 hours isn't bad at all.

But if we could modify cmov etc to work up there it would be alot better and faster I think etc.

Hopefully Mklasson or someone that knows cmov can pick-up on this thread.

[ceselb]

Nbegone isn't listing excluded or out of range factors.
Also mklasson seems to be unrachable at the moment, I sent him a mail over a week ago and still haven't heard back.

But if we could modify cmov etc to work up there it would be alot better and faster I think etc.

Yeah, it will.

Don't make me take off the bag.

[vjs]

ceselb,

If you know something,

I wouldn't mind doing a range up there with cmov, the rate with cmov should be much better than the ~160kps/s I was getting.

Is there some type of switch implemented etc to stop people from sieveing way up there at least I could double check the factor denisty??? Wouldn't take me long to do 3000G.

[ceselb]

The current implementation can't handle those ranges.

[Keroberts1]

is there anyway to improve the speeds of cmov processes for higher ranges? At the speed we're going at we will probably reach the limits of the V.42 before next summer. Of course hopefully we'll find a prime or two by then (I believe we're due to find one soon [dec. 6 ?] ) This is probably best a question for Mklasson so I'll have ot wait for him to find this thread

[vjs]

Don't make me take off the bag.

Are you saying that there is a new sieve client in the works

[ceselb]

No comments.

[Death]

well right now it works at

1714286-1714299 Death
1714299-1714310 Death

1714293 and 1714306 and no factors so far

and there was dnet client running in background. =) now afret stopping it runs much faster.

and is there anybody having mac os x around??

[vjs]

I recieved a message from Mike,

he believes it is an error in the way his calcuations of density work.

He and I also see no harm in testing this range for fun or at least testing the density etc. At this time I'd rather use proth cmov so that additional clock cycles are not lost.

From reading the client thread it looks as though cmov proth will sieve upto 2^60 but anything beyond ~1000T has been "turned off" the client will just stop.

Does anyone have an early version of cmov that will factor this range.

Death have you completed enough of your range to have a decent feel for factor density?

I know I found one factor in 10G which is not bad, but I could have found 2 factors in 1K without testing a large enough range it's tough to tell the density.

[Death]

1714306,5
1714293,5

well, g5 is gone, now I use G4 1GHz dual. speed looks almost the same....

no factors so far =((

well, should wait a few days.

[vjs]

Well death you have reserved a nice little range it's up to you if you'd like to continue. I'd suggest closing the gap,

1714286-1714299 Death
1714299-1714310 Death

1714286-1714310 Death [Complete] and moving on.

I will finish my range once I can get cmov working up there, not much time to work on the G3's right now.

Let me know what happens and how many factors etc.

VJS

[Death]

suxx. no factors at all at this range....

and sf.net have a compile farm, maybe one can try this??

[rogue]

Originally posted by Death

ld: can't locate file for: -lcrt1.o
make: *** [primes] Error 1

does anybody know what all this st*ff means???

I believe that it means that you are trying to staticly link the objects. Edit the Makefile and remove that parameter.

[Grafux]

Hey Death, how'd you get this working on a Mac. I have downloaded the NbeGon_010_osx binary but have no idea what to do with it.

Could you post a step by step so I can start sieving? Thanks.