Mersenne Digest Tuesday, 26 May 1998 Volume 01 : Number 367 ---------------------------------------------------------------------- From: "Michiel van Loon" Date: Sun, 24 May 1998 14:39:31 +0200 Subject: Mersenne: Version 16 upgrade for OS/2 Hi all OS/2 users, To ensure that those using OS/2 won't have to wait until I have solved all porting problems of version 16 I am now releasing V3.1 of PRIMEOS2 and PRIMEEMX. Included are the new LL routines for up to 1024K FFT's, not yet included is the 64 bit factoring code, according to George this is not a direct problem. I hope to make the 64 bit code available as soon as possible, but at the moment it just doesn't go through the tests. Also the more accurate progress reporting to PRIMENET is included in this release. As usual you can find PRIMEOS2 at: http://www.xs4all.nl/~mfvl/prime/ Regards, Michiel van Loon > -----Original Message----- > The server is out of exponents below 5,260,000. Over a 1,000 new machines > have been added to the GIMPS pool lately. I'll work over the weekend > making sure no ranges have fallen through the cracks. ------------------------------ From: STL137 Date: Sun, 24 May 1998 14:45:09 EDT Subject: Mersenne: JLC << < censored :) > >> Thanks, man! =(8-0) STL137 ------------------------------ From: "Scott Kurowski" Date: Sun, 24 May 1998 14:04:59 -0700 Subject: Mersenne: RE: Mersenne Digest V1 #366 > From: matic@cyberia.net.lb (Imad R. Faiad) > Date: Thu, 21 May 1998 17:45:05 GMT > Subject: Mersenne: Is the Primenet Server Down? > > Is the Primenet server down? On and off again for a time, yes. It was back to 100% service yesterday. It needed a bug fix and configuration change to handle a temporary load spike of literally several hundred times the normal transaction rate. Watch the PrimeNet Top (dozen) Producers list over the next several weeks to see why. :-) I only get the digests of this list, so if you have questions about GIMPS PrimeNet, you might email me directly for a faster answer, at primenet@entropia.com. Regards Scott > From: LozCS > Date: Thu, 21 May 1998 20:57:27 EDT > Subject: Mersenne: Update on AOL > > Has anybody got Prime95 to update automatically with AOL? > > I've tried it on 3 separate machines with no joy is it possible? Hi Lawrence, There are several AOL accounts using Prime95 with PrimeNet, some using the RPC version. If you can get IE 4.0 installed and working first, Prime95 will be able to use PrimeNet's HTTP DLL. I've been wrestling with an AOL install problem for someone for over a month, and AOL has been useless, refusing to help their customer install IE 4.0 and refusing to give the proxy info, and of course AOL hides the proxy info, causing the IE4 setup to fail. They had earlier e-mailed a broadcast saying IE4 will work 'out of the box'. The particular dialup AOL server used apparently makes a big difference. Regards Scott > From: George Woltman > Date: Sat, 23 May 1998 14:35:16 -0400 > Subject: Mersenne: Version 16 upgrade > > When upgrading to version 16 be sure you upgrade the primenet.dll too. > If you are using the http protocol, you must copy the new httpnet.dll > on top of primenet.dll The symptom of not doing this is your program will look like a 15.x client to the server, and will be denied LL work when exponents < 5.26M are finally gone. I'll have a stronger version check in the next set of libraries. I'd also like to stress the importance of clearing the pending transactions in the prime.spl file *before* you upgrade the software. regards scott ------------------------------ From: "Rick Pali" Date: Mon, 25 May 1998 09:21:29 -0400 Subject: Mersenne: Locking out factoring assignments on v16.3.5 I've set up my copies of NTprime to 'request numbers to run primality test' exclusively. I *do* understand that if I'm assigned a number that hasn't been factored, my machine will factor it before LLing it. The problem is that is not happening. When my latest number finished this morning, NT prime requested one day worth of work (as I have set it to do) and that consisted of three exponents in the low six-million range. I thought there was some mistake, but when I checked my individual report on PrimeNet, I saw that these were factoring tests. Have I set something wrong, is this a problem with 16.3.5 or has the locking out of factoring been disabled because we're venturing into new territory? Any help would be appreciated. Rick. - ----- rpali@alienshore.com http://www.alienshore.com ------------------------------ From: Eric Brier Date: Mon, 25 May 1998 18:38:29 +0100 Subject: Re: Mersenne: Squareroot of Integers The fastest way to obtain the integer part of the squareroot of n is to compute : (x*x+n*n)/(2*x*n) ---> x until x is stable or makes flip flop between two consecutive values. You can start from any non zero integer. It is quite simple and very fast. E. Brier ------------------------------ From: JT Date: Mon, 25 May 1998 09:50:32 -0700 Subject: WINMAIL.DAT (Re: Mersenne: Re: Mersenne Digest V1 #365) At 07:44 PM 5/21/98 EDT, you wrote: >Uh, what is this DAT file? WINMAIL.DAT? I ain't gonna download this...... >STL137 Winmail.dat is created and sent by a Microsoft email client (I think Outlook and/or Exchange). It contains harmless data, such as text formatting, etc. for other similar MS email clients. It shows up as attachments in non-MS clients. JT ------------------------------ From: "Matthias Müller" Date: Mon, 25 May 1998 20:53:09 +0200 Subject: Re: Mersenne: Locking out factoring assignments on v16.3.5 Did you change your DLLs as well ? http://www.entropia.com/ips/faq.html#outofexponents Rick Pali wrote: > I've set up my copies of NTprime to 'request numbers to run primality test' > exclusively. I *do* understand that if I'm assigned a number that hasn't > been factored, my machine will factor it before LLing it. > > The problem is that is not happening. When my latest number finished this > morning, NT prime requested one day worth of work (as I have set it to do) > and that consisted of three exponents in the low six-million range. I > thought there was some mistake, but when I checked my individual report on > PrimeNet, I saw that these were factoring tests. > > Have I set something wrong, is this a problem with 16.3.5 or has the locking > out of factoring been disabled because we're venturing into new territory? > > Any help would be appreciated. > > Rick. > ----- > rpali@alienshore.com > http://www.alienshore.com ------------------------------ From: George Woltman Date: Mon, 25 May 1998 15:07:00 -0400 Subject: Re: Mersenne: Locking out factoring assignments on v16.3.5 Hi Rick, At 09:21 AM 5/25/98 -0400, Rick Pali wrote: >[Rick asked for LL tests only]...three exponents in the low six-million range. I >thought there was some mistake, but when I checked my individual report on >PrimeNet, I saw that these were factoring tests. > >Have I set something wrong, is this a problem with 16.3.5 or has the locking >out of factoring been disabled because we're venturing into new territory? The server is once again out of exponents below 5.26M to assign. That means version 15 users will get factoring assignments until they upgrade to version 16. As to why you are seeing this problem in version 16, I do not have an answer. Scott and I will figure it out. More than 1500 exponents above 5.26M have been successfully assigned for LL testing to version 16 clients, so I don't have any ideas right now. Regards, George ------------------------------ From: Chris Nash Date: Mon, 25 May 1998 20:39:30 -0400 Subject: Re: Mersenne: Squareroot of Integers Eric >(x*x+n*n)/(2*x*n) ---> x until x is stable or makes flip flop between >two consecutive values. You can start from any non zero integer. It is >quite simple and very fast. This sort of looks a bit odd, particularly as x=sqrt(n) isn't a fixed point of it. I think what you were aiming for was equivalent to Newton's method: f(x)=x^2-n, so iterate x-f(x)/f'(x) <- x, which is (x*x+n)/(2*x) <- x. Not a bad iterative method, and the error halves every iteration, ie you get one extra bit every iteration. There are faster methods though for n with large numbers of bits, given that you will have a system square root function for 'native' floating-point numbers. If you express n as n=(2^2k)(a^2)(1+x), where 1<=a<2 is chosen so that it has h bits, half the number of bits of a 'native' number on your system (basically compute a from the C library or system squareroot of the leading bits of n, and truncate it to h bits so that a^2 fits in such a 'native' representation). By this method x will be very small, around 1/(2^h), and so the terms of the binomial expansion of (1+x)^.5 decay as x. Each term then will give you about h bits of accuracy, you need about k bits to get your root accurate to an integer, and your square root is (2^k)a(series expansion). n/(a^2) should have more than enough bits for you to get a good value for x. Chris Nash ------------------------------ From: Brian Beuning Date: Mon, 25 May 1998 21:14:24 -0400 Subject: Re: Mersenne: Locking out factoring assignments on v16.3.5 I noticed a little unexpected behavior when we switched from version 15 to 16. Each machine had two exponents checked out, call them A and B. On version 15, they had been crunching away on the LL test for A. When we switched to version 16, they started trying to factor B. When they were done factoring B they resumed the LL test on A. Then I assume it is going to do the LL test for B. Brian George Woltman wrote: > Hi Rick, > > At 09:21 AM 5/25/98 -0400, Rick Pali wrote: > >[Rick asked for LL tests only]...three exponents in the low six-million > range. I > >thought there was some mistake, but when I checked my individual report on > >PrimeNet, I saw that these were factoring tests. > > > >Have I set something wrong, is this a problem with 16.3.5 or has the locking > >out of factoring been disabled because we're venturing into new territory? > > The server is once again out of exponents below 5.26M to assign. > > That means version 15 users will get factoring assignments until they > upgrade to version 16. > > As to why you are seeing this problem in version 16, I do not have > an answer. Scott and I will figure it out. More than 1500 exponents > above 5.26M have been successfully assigned for LL testing to version > 16 clients, so I don't have any ideas right now. > > Regards, > George ------------------------------ From: Eric Prestemon Date: Mon, 25 May 1998 18:19:20 -0700 (PDT) Subject: Mersenne: Linux 16.x? We Linux mprime users are still at 15.x, and apparently will not be able to check out exponents from now on. Are there plans/effort to update the linux client to the 2*10^7 max exponent of the 95/NT 16.x clients? I don't recall seeing anything here. - -Eric (Note, this is not a complaint, just a query. I'm thrilled there's any Linux client) ------------------------------ End of Mersenne Digest V1 #367 ******************************