Offline SNFS

[R.D. Silverman]

Hello All!

If any of you have machines that are offline (or not constantly on-line)
and would like to help out with SNFS, I can use the help. I will provide
source (either Unix or Windows) as well as data if anyone wants to help
out.

I am about 75% done with 2,661+, and will then do 2,1238L and 2,667+.

I can also provide an SGI binary (the only Unix system I have) or a Pentium
IV binary. However, the source will let people see how to implement SNFS.

Paul: I am not trying to poach. These are machines not accessible to NFSNET
since they are offline.

Bob

[S00113]

Originally Posted by Bob Silverman

If any of you have machines that are offline (or not constantly on-line)
and would like to help out with SNFS, I can use the help. I will provide
source (either Unix or Windows) as well as data if anyone wants to help
out.

I need a suitable test case for jobs in the Condor http://www.cs.wisc.edu/condor/ standard universe. See this link for limitations on jobs in the standard universe: http://www.cs.wisc.edu/condor/manual...mitations.html

Is it a candidate?

[R.D. Silverman]

Originally Posted by S00113

I need a suitable test case for jobs in the Condor http://www.cs.wisc.edu/condor/ standard universe. See this link for limitations on jobs in the standard universe: http://www.cs.wisc.edu/condor/manual...mitations.html

Is it a candidate?

I know about Condor. Yes, SNFS would be an excellent test case. My code
can either have each machine run from its own private directory, or
multiple machines can all run from a single directory (which is the recommended approach).

Bob

[S00113]

Originally Posted by Bob Silverman

I know about Condor. Yes, SNFS would be an excellent test case. My code can either have each machine run from its own private directory, or multiple machines can all run from a single directory (which is the recommended approach).

Where can I find source and instructions?

[junky]

im really interested into that subject Bob.
My twin has a P4-2.8-800FSB, which can help a bit.
I've 2 p2-300 at my parent's house which will can run the client and upload WUs like once a week or once/2 weeks.

Sometimes, our network going down, so every machines are going into idle, so placing a "module" for offline machines is a AMAZING addon.

my old brother has dialup, so he'll be able to help too. Like many people using dialup or DSL, they'll be able to sieve until they connect to the net.



nice job Bob.

[Wacky]

Originally Posted by Bob Silverman

If any of you have machines that are offline (or not constantly on-line) and would like to help out with SNFS, I can use the help.
<snip>
Paul: I am not trying to poach. These are machines not accessible to NFSNET since they are offline.

Don't be fooled! He's trying to invade our NFSNET participant base.

But Bob is a "good guy"(tm). Remember that it was HIS siever that I modified to use in the original NFSNET many years ago.

Bob's agenda is somewhat different from mine. He is concerned with parameter selection and similar mathematical properties of the NFS method. I'm more interested in coordinating the computations.

From my perspective, the way that he is doing things suits his purpose, but has two drawbacks.

First, it doesn't scale. He can effectively use the help of, perhaps, a dozen participants. But the administrative overhead will swamp him at a few hundred. (Been there, tried that)

Secondly, by providing an alternative, he diverts developers from helping NFSNET solve the problems which are making it difficult for those machines to participate in our infrastructure.

OTOH, I haven't been all that successful in getting those developers to actually produce anything useful. (Yes, you know who you are. Although you promised to deliver, I'm still waiting)

So if you have some machines that, for whatever reason, cannot currently participate in the NFSNET sievings, please help Bob.

And if you have some programming skills and time, please help me develop the next generation of the NFSNET software.

I hope that we will eventually develop the system into one that allows you to run Bob's siever either "stand alone" or as a part of the NFSNET infrastructure.

We all share what we learn. So help out where you can.

[R.D. Silverman]

Originally Posted by Wacky

First, it doesn't scale. He can effectively use the help of, perhaps, a dozen participants. But the administrative overhead will swamp him at a few hundred. (Been there, tried that)

Agreed. I've been there as well. But I only expect a few responses. When
you only have 5 machines, 5 more doubles your speed. When you have
200, 5 more only has small effect..

Originally Posted by Wacky

Secondly, by providing an alternative, he diverts developers from helping NFSNET solve the problems which are making it difficult for those machines to participate in our infrastructure...

Please. NFSNET does not allow for offline computing and Paul told me that
you have given up on the idea. I explicitly called for off-line help. I am
not "poaching".

[Wacky]

Originally Posted by Bob Silverman

Please. NFSNET does not allow for offline computing and Paul told me that you have given up on the idea. I explicitly called for off-line help. I am not "poaching".

The current implementation of NFSNET does not handle off-line and we have given up on attempting to try to make it do so.

However, off-line computing is one of the goals for the next generation. So is having an, optional, GUI for all platforms. In addition, I expect to be able to support multiple sievers for NFS and perhaps other DC tasks such as polynomial search for gnfs, ecm, etc. all within the same framework.

But, instead of writing that new code, I've had to worry about replacing computing resources that are soon going to become unavailable to us, etc.

Note to Bob: I never said that you were "poaching". In fact I said

So if you have some machines that, for whatever reason, cannot currently participate in the NFSNET sievings, please help Bob.

[xilman]

Originally Posted by Wacky

The current implementation of NFSNET does not handle off-line and we have given up on attempting to try to make it do so.

Thanks for the clarification. When Bob mailed me I did, indeed, tell him that we'd halted the experiment of using off-line sievers with NFSNET because the human effort on behalf of both contributors and admins was just too great. That experiment necessarily used the current implementation of NFSNET!

I don't consider Bob to be poaching either. If he can use a resource we can't, good luck to him.

Paul

[junky]

of course, if he can help with next generation that would be really great.

Starting many linesiever for a simple machine, that would be great too.

But Richard, what's the estimated time required for these tasks (coding and testing) ?

[xilman]

Originally Posted by junky

of course, if he can help with next generation that would be really great.

Starting many linesiever for a simple machine, that would be great too.

But Richard, what's the estimated time required for these tasks (coding and testing) ?

Is that an offer of help?

Richard is the person best placed to answer your question, but one response must surely be that the time required depends greatly on the amount of effort put into the tasks.


Paul