Distributed GGNFS on Windows
 

Hi guys,

First off, if the moderator of this category finds that this topic can be merged with the topic "GGNFS Question & Answers", I have no objections to it. :bow:

For I start, I must say that I claim no responsibility for damage, security risks or viruses.

I am assuming that you may use GGNFS & assorted programs on your pc's.

First, you have n-boxes with Windows (from Windows 2000 and up I assume). (Duh)

On the first PC, install the ggnfs suite
Modify the Perl as following:

[code]
.....
$GGNFS_BIN_PATH=".";
....
$SYS_BIN_PATH=".";
.....
[/code]

Remeber, this is Windows not Linux.

Share this directory (rigth click on tha folder) & tick the option "Another user can modify" this folder.
Or give the other PC "full control" of this folder. This can be seen as a possible security risk.

On the second PC, map the Network drive (\\pc1\ggnfs) to a "normal" drive letter. ("M:\")

On the first Pc, fire up the GGNFS Suite, like

[code]
c:\ggnfs\factlat.pl c97.poly 1 x
[/code]

where x equals the number of nodes you want deploy the factorization on it.

Then, on Box n, drop to the command point:

[code]
Start => cmd.exe (or the equivalent)
[/code]

type in:

[code]
M: (change to "drive" M)
[/code]

And finally:

[code]
M:\factlat.pl c97.poly 2 2
[/code]

Please make sure, that you have enough bandwith and that your router // hub stays online.

And off you go!!! :banana:

BTW, I am running my Distributed run on a "normal" P4 3 Ghz & in a VmWare Windows Xp Pc. :geek:

PS. Sorry that my tone seems so nervous but for me it is a big leap!!! :smile: :smile: :smile:
PPS. BTW, I know something about Networking :showoff: :flex: and Windows :devil: :geek:

I have never ran GGNFS, but there is a question for you.

Is it possible that the congruence relation file corrupts because both processes (local and remote) try to write to this file?

No, I believe each client writes raw relations to a different file, and the host client " factlat.pl number.poly 1 n " is the only one that processes these relations.

[QUOTE=alpertron]I have never ran GGNFS, but there is a question for you.

Is it possible that the congruence relation file corrupts because both processes (local and remote) try to write to this file?[/QUOTE]

Diego,

:shock: I don't know exactly what you mean, but I try to give you a decent answer.

Several files has the following @ the end: .x (where x == the number of the client). IMHO, IIRC, GGNFS (:wink:) "adds" the relations collected to the spairs.add.x file & after the processing of relations is done on the main CPU, (please correct me if I am wrong) the relations collected into the spairs.add.x file are added into the main run.

Following files seems to have the .x extension:

[code]
spairs.out2
c97.job.2.afb
.last_spq2
c97.job.2
[/code]

Please, if anyone knows better, please step in & fill the blanks...

Regards,
:coffee: :coffee:

Please,

Also remark that you [b]CANNOT[/b] use (yet) [b]factLat.pl xxx.N 1 x[/b].
Apperantely, there is no distributed polysel implemented yet. (I think)

Apperantely, only the distributed part is "distrubted" as of now (I think)

Regards,
Cedric Vonck

VJS, if that is the case, why don't you run the program in both computers on drive C: and then merge the results after getting all relations? In this way you don't need a permanent connection.

[QUOTE=alpertron]VJS, if that is the case, why don't you run the program in both computers on drive C: and then merge the results after getting all relations? In this way you don't need a permanent connection.[/QUOTE]

Diego,

For these kind questions, please ask them to Chris Monnico.
I am only [i]try[/i] to understanding the basics of SNFS....
:surrender :surrender

Regards,
C.

OK. Notice that my name is Dario, not Diego.

[QUOTE=alpertron]OK. Notice that my name is Dario, not Diego.[/QUOTE]


Oops :redface: :redface:

[QUOTE=alpertron]VJS, if that is the case, why don't you run the program in both computers on drive C: and then merge the results after getting all relations? In this way you don't need a permanent connection.[/QUOTE]

Sure it could be done that way the only problem is that's not the way the perlscript is designed.


I've never used the program over a network ... I've only used it on a dual processor machine. From what I've gathered in my mucking around with "snfs only". Using the perlscript and two processors, the perlscript creates two job files, one dedicated to each processor.

Each processor sieve over a range of q, everytime that range is finished the perscript checks for the next highest q starting point. Only when processor 1 completes a sieve range does it process the relations and determine if more sieving is required.

This is good and bad: Bad b/c the perscript runs procrels far too often, and when enough relations are found the script won't stop the other processor.

This isn't really the way ggnfs or snf should be implemented.

The steps should be:
1. determine a good polynomial, alim, rlim, q, range in which q is sieved over to produce enough relations (maybe 5-10% more).
2. Divide that range of q over several machines, and sieve it all.
3. Merge the sieve data
4. Process the relations...
5. determine if you have enough...
6. More sieve or prune etc...
...
...
...
Sovle the matrix.

Problem is the script is very simple and does everything for you... the trade off is it doesn't do step 2 and does step 4 far too often. It simply process the relations every x number of q trying to create a matrix as fast as possible. Alot of time is wasted in the processing of relations and trying to create a matrix step.

Give it a try you'll see what I mean, the script handles numbers <110 digits quite easily but for 140+ digit... You need a little more than the script.

[QUOTE=VJS]This isn't really the way ggnfs or snf should be implemented.

The steps should be:
1. determine a good polynomial, alim, rlim, q, range in which q is sieved over to produce enough relations (maybe 5-10% more).
2. Divide that range of q over several machines, and sieve it all.
3. Merge the sieve data
4. Process the relations...
5. determine if you have enough...
6. More sieve or prune etc...
...
...
...
Sovle the matrix.

Problem is the script is very simple and does everything for you... [/QUOTE]I must be missing something.

Surely, as you have the full source of the script and as you have a precise description of its failings and how to remedy them, why don't you just modify the script?

Paul