Prime95 version 25.7

[garo]

Thanks.

[ET_]

Originally Posted by monst

Factor5 multi-threads TF. Luigi, do you care to comment...

I don't know well how Prime95 factoring code works. If it runs over 16 residual classes then yes, it may easily be susceptible of multi-threading.

The problem may be on memory contention among cores and processors; also you may need to rewrite (and deeply retest) part of the code, check interactions with the exiting multi-threaded code, define some user-defined threading policies... Trying to modify an already hyperoptimized algorithm it's not a piece of cake, and the GIMPS team has some heavy tuning work to do on the server side right now.

I'm sure George is already thinking at it.

Luigi

[Prime95]

Originally Posted by ET_

I don't know well how Prime95 factoring code works. If it runs over 16 residual classes then yes, it may easily be susceptible of multi-threading.

I'm sure George is already thinking at it.

It is not technically difficult to do. It would just take some coding time and testing.

However, there is little reason to do it. One could simply set up two workers each doing TF. The only advantage I see is that when a sstem setup to do multi-threaded LLs (not the default) does get an LL assignment that needs TF, then it has an idle core.

[ET_]

Originally Posted by Prime95

It is not technically difficult to do. It would just take some coding time and testing.

However, there is little reason to do it. One could simply set up two workers each doing TF. The only advantage I see is that when a sstem setup to do multi-threaded LLs (not the default) does get an LL assignment that needs TF, then it has an idle core.

There is another reason to use multi-threaded TF, but of course it is outside GIMPS realm: LMH would benefit it to quicker finish deep factoring on single exponents using dual and quad cores, or pushing the limit one or two bits further.

Luigi

[James Heinrich]

Originally Posted by Prime95

The only advantage I see is that when a sstem setup to do multi-threaded LLs (not the default) does get an LL assignment that needs TF, then it has an idle core.

One idle core for a few hours doesn't seem significant; over the next 1-2 years today's 4-core CPUs will evolve into 6-, 8-, and 12-core CPUs and it makes less and less sense to have single-threaded work that can't take advantage of available processing power. Granted it's easier to leave the status quo, and hope the average user won't alter the default settings, but TF is (as far as I know) the only major worktype that is single-threaded. I'm sure I'm a statistical outlier, but I'd love to be able to power through TF to 2^79 in days rather than weeks on my octocore machine. I know this is George's project and it's entirely up to his discretion where to spend his coding resources, but I submit this would be a useful investment of whatever hours it takes.

[S00113]

I tried running mprime 25.7 on some old rubbish^W computers used as thin clients, and it detects two CPUs or cores on AMD Duron.

processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 7
model name      : AMD Duron(tm) processor
stepping        : 0
cpu MHz         : 1000.057
cache size      : 64 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow up ts
bogomips        : 2002.30

[Prime95]

Originally Posted by S00113

I tried running mprime 25.7 on some old rubbish^W computers used as thin clients, and it detects two CPUs or cores on AMD Duron.

Here is the C code:

	count = 0;
	fd = fopen ("/proc/cpuinfo", "r");
	if (fd != NULL) {
		for ( ; ; ) {
			if (fscanf (fd, "%s", buf) == EOF) break;
			buf[9] = 0;
			if (strcmp (buf, "processor") == 0) count++;
		}
		fclose (fd);
	}
	if (count == 0) count = 1;
	return (count);

Basically it counts lines in /proc/cpuinfo that begin with "processor". What does your /proc/cpuinfo file contain? If it is the text in your last message then I don't see how count became two.

[James Heinrich]

Originally Posted by Prime95

If it is the text in your last message then I don't see how count became two.

Originally Posted by S00113

Maybe?

[garo]

Prime95 is suddenly unreserving exponents for no rhyme or reason. I did not unreserve the exponents either from Prime95 or from the web page. And I just checked the exponent status and the exponents are not assigned to anyone else a full three days after they were removed from my worktodo. Here is the relevant bit from prime.log:

[Fri Nov 21 13:55:05 2008 - ver 25.7]
Unreserving M18259193
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=au&g=7c2b4ed4319e777115f4e8d5b96faef4&k=3A0F4B91114957D3C13D42962DEDA7C0&ss=10306&sh=2CE08B4413AEFB153BBE7165C059A483
RESPONSE:
pnErrorResult=0
pnErrorDetail=SUCCESS
==END==

Unreserving M18357149
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=au&g=7c2b4ed4319e777115f4e8d5b96faef4&k=16AC7518C57104A3BCE2712981AD7196&ss=20659&sh=9FAB59D9AB30BB0F9A9F9E346A1FE35B
RESPONSE:
pnErrorResult=0
pnErrorDetail=SUCCESS
==END==

Unreserving M18370567
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=au&g=7c2b4ed4319e777115f4e8d5b96faef4&k=F10CA810A8A8444708A90414F38DF628&ss=23044&sh=B42AD571BB8B8AEC9DE540D96C806365
RESPONSE:
pnErrorResult=0
pnErrorDetail=SUCCESS
==END==

Getting assignment from server
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=ga&g=7c2b4ed4319e777115f4e8d5b96faef4&c=0&ss=41&sh=10C78AAD38BB7BDDDA40F6CD1C48AD5F
RESPONSE:
pnErrorResult=0
pnErrorDetail=Server assigned Lucas Lehmer primality double-check work.
g=7c2b4ed4319e777115f4e8d5b96faef4
k=47AB1D1C4E39CE5374D77CA9A1B1CCC8
A=1
b=2
n=23182069
c=-1
w=101
sf=67
p1=1
==END==

PrimeNet success code with additional info:
Server assigned Lucas Lehmer primality double-check work.
Got assignment 47AB1D1C4E39CE5374D77CA9A1B1CCC8: Double check M23182069
Sending expected completion date for M23182069: Dec 04 2008
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=ap&g=7c2b4ed4319e777115f4e8d5b96faef4&k=47AB1D1C4E39CE5374D77CA9A1B1CCC8&c=0&p=0.000000&d=86400&e=1150067&ss=18467&sh=07475AC47472088D09E6BA960163BBDA
RESPONSE:
pnErrorResult=0
pnErrorDetail=SUCCESS
==END==

Getting assignment from server
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=ga&g=7c2b4ed4319e777115f4e8d5b96faef4&c=0&ss=6334&sh=86EA894EE18A34B5F9996D5014F6912A
RESPONSE:
pnErrorResult=0
pnErrorDetail=Server assigned Lucas Lehmer primality double-check work.
g=7c2b4ed4319e777115f4e8d5b96faef4
k=B5224D10CD65FA47986631FA0FA6ABEA
A=1
b=2
n=23182123
c=-1
w=101
sf=67
p1=1
==END==

PrimeNet success code with additional info:
Server assigned Lucas Lehmer primality double-check work.
Got assignment B5224D10CD65FA47986631FA0FA6ABEA: Double check M23182123
Sending expected completion date for M23182123: Dec 14 2008
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=ap&g=7c2b4ed4319e777115f4e8d5b96faef4&k=B5224D10CD65FA47986631FA0FA6ABEA&c=0&p=0.000000&d=86400&e=1965520&ss=26500&sh=7F57559CC641B1C42AC270DDD59A0B71
RESPONSE:
pnErrorResult=0
pnErrorDetail=SUCCESS
==END==

I just manually added these exponents back in and they are now on my assignments page and registered in my worktodo. But I was able to unreserve the two 23Ms without using the web interface. I don't know if that is a bug.

Edit: The exponents had just updated their completion date less than an hour before they were unreserved. And it is definitely not a "Days of Work" issue.

[starrynte]

I got one random unreserve too

[Fri Nov 21 13:17:48 2008 - ver 25.7]
Unreserving M18817921
URL: http://v5.mersenne.org/v5server/?v=0...2154B766E60E88
RESPONSE:
pnErrorResult=0
pnErrorDetail=SUCCESS
==END==

though i too managed to re-reserve it later

[jinydu]

Why even add such an instruction (unreserving) to the client? Wasn't the previous system of the server and manual communications doing all the unreserving good enough?