Welcome to Seventeen Or Bust

[IronBits]

www.SeventeenOrBust.com :D
What's Seventeen Or Bust?
SB ("Seventeen or Bust") is a distributed attack on the Sierpinski problem. The problem deals with numbers of the form N = k2n + 1, for odd k and n > 1. Numbers in this form are called Proth numbers. If, for a particular value of k, every possible choice of n results in a composite (non-prime) Proth number N, that number k is called a Sierpinski number. The Sierpinski Problem asks "what is the smallest Sierpinski number?"

John Selfridge proved, 40 years ago, that k = 78,557 is a Sierpinski number. It is generally believed that this is the smallest, but it hasn't yet been proven. To prove it, we need to find, for every k less than 78,557, some number n such that the Proth number N is prime. This has already been done for all but seventeen values of k: 4847, 5359, 10223, 19249, 21181, 22699, 24737, 27653, 28433, 33661, 44131, 46157, 54767, 55459, 65567, 67607, and 69109. We hope to break at least some, and hopefully all, of these remaining numbers.

[ColinT]

IB: Can you say that again with words I can understand?

[ColinT]

OK, I liked SOB when it first came out but I had a LOT of trouble getting ahold of the people running the project. They have no forum. Maybe Free-DC can fill this void.

There is a new client out now. It runs 30 times faster. However the work unit run time is gone from 45 minutes to 8 hours on a fast box. And there is no checkpointing. We really need checkpointing.

To sum it up, you don't need to know anything about Sierpinski Primes. It's another math project that is fairly easy to rise to the top of the stats.

The client hides away nicely and does not hog your system

This client needs a FAST CPU. P500s need not apply.

There are clients for several OSes now. SO let's rejoice and beat dp and his horde.

[FoBoT]

thanks for the post ColinT
i saw the "30x" faster thing last night and re-installed, then when i get home from work today i see it had only done 3 blocks on my 1.4Ghz T-bird! :scared:

i thought something was wrong, your explanation tells the tale ;)

[ColinT]

Now they are searching the whole "Keyspace", not just small slices. So, yes the client runs 30 times faster but it has 10 times more space to search.


It's an 8 hour WU on very fast boxes. Let's see what you have:)

Three blocks a day is normal. You will rise to the top quickly with three a day.

[Darkness Productions]

Heh... my p3 500 (here at work) slips through a full WU in way to long (like almost 24 hours or so).

Speaking of which, can we get people to post approximate cEMs/sec and their system setup here? I want to know which processor setup is the fastest ;)

This p3 gets a little over 22k cEMs/sec...


Originally posted by ColinT
This client needs a FAST CPU. P500s need not apply.


Hey now.... I'm on your side, but I' want to at least say that I've beaten you in at least one project :notworthy :notworthy



There are clients for several OSes now. SO let's rejoice and beat dp and his horde.

[ColinT]

DP:

We already know what the fastest CPU is. And it's not a P4. Even Tbirds scream, but the clear winner is an AMD XP. I don;t have a new one yet, but this XP1700 just started a new block. 1% in 8 minutes:)

cEMs/sec is still rising now, but it's at 65,100

Update: 10% in 48 minutes

[ColinT]

I've had a ton of failures over the last three days. The client gets up to about 90% done and stops. And I had a crash once.

This client is buggy.

[ColinT]

SOB will not complete a WU any more on my P4 and my XP1800+. It looks like this:

http://www.zweb.com/log/frozen_sob.jpg

I'm down to a single 1400 TB left.

[Darkness Productions]

Colin - Maybe an uninstall/reinstall is in order? Or have you already tried that?

[ColinT]

It worked for a short time - a day - and then never worked again. We need v.93

[ColinT]

The servers are back up. This was posted on their web page:




8/23: Downtime
Lost power for about two hours this evening - unfortently the power backup only lasted about 45 minutes. Sorry for the downtime, everything should be okay now.

[FoBoT]

Originally posted by ColinT


This client is buggy.

mine crashed too (XP)

[ColinT]

We need to get ahold of the programmer and give him these bug reports and suggestions. I'll try when they get back from vacation.

[Darkness Productions]

Well, I restarted my clients (4 of them) and I'm gaining on the ColinTATOR. He's a single spot ahead of me, and I'm hoping to catch him very soon........:cheers:

[ColinT]

I have two boxes running now. I hope to maintain my lead for at least a day:) I just can't seem to keep the clients running.

[Darkness Productions]

I finally passed Colin!!!! My lead is very touchy at the moment, but as soon as a couple of other boxen dump, it'll be all gravy from there on. Now, on to the #1 spot.

[ColinT]

Whooops. There went your lead:) I have all three boxes on it now. #2 daily and going up fast. 2 boxes 24/7 and one about 14/7.

Catch me if you can!

[ColinT]

Relative speeds:

P4 2000=5400 cEMs/sec
1400 Tbird = 15000 cEMs/sec
xp 1700+ = 55000 cEMs/sec

Anyone else?

[DuffMan|TSF|]

axp 1900+ = 84K cEM/sec

[ColinT]

I've noticed something. SOme of the WUs - on my XP1700+ - take from 4:08 to 12:00 plus. It will run off a string of fast ones and then slow down to to a near crawl. Anyone else notice this?

[jjjjL]

Colin - a wu shouldn't vary in time that much. could you check your log and email me the timestamps showing this behavior?

Thanks,
Louie

[ColinT]

Yes, I will do that. For several days last week I was getting very fast WUs. Now they have reverted to 12 hours.

I thought I was running the older client, but after a check, I found I am not.

[ColinT]

I popped in a new mobo and now my XP1700+ is running at 83157 C/m/secthingies.

Imagine my new Barton:-)

[jjjjL]

I have been working on SB this weekend. I tried to add in some of the ideas I saw here.

Features of the soon-to-be-available v0.9.6:

-Checkpointing works! Saves progress every 10 minutes. :D

-Config can be modified while program runs. [colin asked and so he recieves]

-Fixed the only known "bug" in v0.9.2 which was the cEM/s dropping after a couple days due to an overflow. The resolution has been reduced a little for increased performance. I'll find a balance by v1.

-Removed sec/blk and the avg. net since both were inaccurate and confusing to users... you have no idea how many people told me SB had "trashed their net connection" because it said 2b/s in the net avg. :rolleyes:

-Removed Realtime/High priority from default options. Some users (like my little brother ;)) crashed their computers rather successfully in an attempt to make the program run faster. (thanks anyway Pete :))

-Progress meter now has 100x as much resolution! (i.e. 65.47% instead of 65%) I really like it.

-Modified the timer startup code slightly so as to make the run time start when the block starts processing instead of when the program stops. This will make cEM/s more stable (instead of the current upramp on startup). Dave and I have actually disagreed about which is more accurate. Let me know what you think.

-Client now warns on Exit to make sure user intends to close SB. [again, colin's idea.. thanks.]

-Did I mention that caching works!! :D


OK, so I'm testing the client myself right now. It's performing flawlessly so far on test data but I want it to do real blocks on the server so it's taking time (67.33% done right now on a couple computers). I'll post the new version on the website tomorrow or sometime this week if all goes well.


I just looked over some of dp's ideas which aren't in v.94. Here are my thoughs:

service install - I want to do this too. On the TODO list.

external config - sounds like a good idea. would make it easier to deploy on multiple machines. i'll have to make that available when team support is done.

seperate cores - Non issue. The core has modified asm routines for every x86 instruction (including P4 SSE2).


that's my news. these are my thoughts. let me know what you guys think.


Later --
Louie

[ColinT]

Hey Doc!!! This is great! I am looking forward to testing this as soon as tomorrow. I'll hit IRC and warn the others to look for it.

Caching a great feature. It opens up a vast new area to explore.


-Fixed the only known "bug" in v0.9.2 which was the cEM/s dropping after a couple days due to an overflow. The resolution has been reduced a little for increased performance. I'll find a balance by v1.

I notice this on the slow boxes. Like my P4 2000. It's run 24 WUs straight now. The Cems are down to about 3000 now.

As to when the program starts calculating, I have no idea - really.

I've been building boxes lately. Checkpointing will really help me. I think I have 22 outstanding WUs now:) Ooops.

Thank you for the heads up. You guys are tops!

[ColinT]

Couple more thing I have been thinking about:

After the client has run enough to determine it's speed, put up a projected finish time.

Record the WU start and stop times in the log so we can write a monitor program. All we need is a couple of numbers to parse.

This is a sample of monitoring programs:

http://www.plasticstudios.org/robsdelphipage.htm

One last thing:) There is a block that tells how many seconds has elapsed on the current block. Can you change the format to h:m:s?

I can't tell time very well with seconds:)

[FoBoT]

hey sounds good, nice timing too

the DF protein that comes up tomorrow should be quite slow, i'll switch all my home boxen to SB and help test out the new client :)

[jjjjL]

colin - the log already has timestamped messages showing where it starts and stops.

what other information would be helpful? if everyone on the forum can help develope a better log format, I'd be very receptive to changing the way it's done. please be specific... don't give me a: Do it like SETI does or anything like that. ;)

right now it looks like this:

[Sun Aug 04 21:49:30 2002] got proth test block from server
[Sun Aug 04 21:50:12 2002] connecting to server
[Sun Aug 04 21:50:16 2002] logging into server
[Sun Aug 04 21:50:16 2002] requesting a block
[Sun Aug 04 21:50:21 2002] got proth test block from server
[Sun Aug 04 21:52:15 2002] completed proth test block: result 3


perhaps adding k/n values to completion / retrieval messages?

your suggestions are hearby welcomed. :)

-Louie

[jjjjL]

my test block worked w/ no problems. i'm gonna post v0.9.4

i like colin's eta idea but i have to just get this out.

keep giving me feedback. i'll make the linux client next weekend if i don't find time this week.

[jjjjL]

client is posted. let your current blocks finish if you can then grab this babe. http://www.seventeenorbust.com/ has all the info.


i'm going to sleep so i can make it to class tomorrow folks ;)


nite, i'm out.

-Louie

[alpha]

Wow, that's a pretty nice client :)

Just a couple of points. It might be nice to have the progress percentage popup when you hover the pointer over the systray icon.

What are 'Retries' and 'Rprt tries' in the config? I'm assuming its something to do with retrying to get a WU if an attempt fails?

If a machine is offline when a WU is completed, what would happen? Would it keep attempted to send back results and fetch a new WU, and keep failing? How about some kind of local WU caching/queueing, for those with non-permanent connections?

I've only been using 0.9.4 for ten minutes, so apologies if I've requested something that already exists :p

[ColinT]

I'm having trouble with my XP1700. THe client starts and gets to someplace less than 1% and then does not advance. The time counter works, and the cEM/s moves.

I'll go try it on my P4 now.

Looks like it's working on the P4.

[jjjjL]

colin - can i get a screen shot of that?

-L

[ColinT]

http://www.lapco.net/sob_stopped.jpg

I have one more AMD box I can try it on in 15 minutes.

EDIT: That 1400 Tbird works fine. Only my XP1700 fails.

[jjjjL]

colin - the XP worked w/ v0.9.2 but not v0.9.4?

is there any reason why it wouldn't be able to create/read/write a cache file in the directory you're runing from? you aren't running it off a CD or something like that?

-L

[ColinT]

Yes, it worked fine with .92. I can't think of a reason. There is plenty of disk space. The cache file is not written. Here's the log:

[Mon Sep 09 08:11:59 2002] connecting to server
[Mon Sep 09 08:12:00 2002] logging into server
[Mon Sep 09 08:12:00 2002] requesting a block
[Mon Sep 09 08:12:01 2002] got proth test from server
[Mon Sep 09 08:24:04 2002] restarting saved proth test
[Mon Sep 09 08:32:29 2002] restarting saved proth test
[Mon Sep 09 08:45:11 2002] restarting saved proth test
[Mon Sep 09 09:19:01 2002] restarting saved proth test
[Mon Sep 09 10:55:47 2002] restarting saved proth test
[Mon Sep 09 10:56:46 2002] restarting saved proth test
[Mon Sep 09 10:57:27 2002] restarting saved proth test

[ColinT]

I put eCCP on that XP1700. Runs fine. I am at a loss.