Internet proof of R109297 feasibility study

[bur]

How long do you expect it to take? I have an i9-10920 with 12 cores and static IP that could be used.

ecpp-mpi is working.


edit: Regarding "how long" I just saw your estimation from the first post. So we need 400-500 cores to make this managable, right?

[paulunderwood]

Originally Posted by pinhodecarlos

So why have you said the limiting factor was the net.

The limit is the amount of data we can squeeze down the net from the master. Several slaves might be okay but hundreds of them might be a big problem.

[paulunderwood]

Originally Posted by bur

How long do you expect it to take? I have an i9-10920 with 12 cores and static IP that could be used.

ecpp-mpi is working.


edit: Regarding "how long" I just saw your estimation from the first post. So we need 400-500 cores to make this managable, right?

Have you got GWNUM working with ecpp-mpi?

Will you be willing and able to give me an account with ssh access?

If the answer is yes to both of these questions then pm me with details.

An initial trial will be on a "small number". In the pipe-line for me I have an end of May Mersenne cofactor I am working on. We could smash through that; or we could test the system with a smaller number.

I estimated 100k core days but that might well be exaggerated. More cores is better, but there will be diminishing returns if slaves fall over and stop the whole system.

[bur]

Originally Posted by paulunderwood

Have you got GWNUM working with ecpp-mpi?

You mean by following your instructions for building ecpp? I did that, but it's not up-to-date, I built it Jan 10. So I'd better build an updated version prior to testing.

Will you be willing and able to give me an account with ssh access?

Yes, what rights does that user need? Is access to a folder containing the ecpp-mpi binary sufficient?

The system is setup to accept authentication via SSH keys only, will that work? I could change it, but would strongly prefer not to.

If the answer is yes to both of these questions then pm me with details.

I kept these general question here, because it might be interesting to others as well. Also I will need a while to set everything up, building ecpp with the various patches took some time. I think some time next week. Maybe in the meanwhile others will join as well.

An initial trial will be on a "small number". In the pipe-line for me I have an end of May Mersenne cofactor I am working on. We could smash through that; or we could test the system with a smaller number.

I estimated 100k core days but that might well be exaggerated. More cores is better, but there will be diminishing returns if slaves fall over and stop the whole system.

Sounds very good.

[paulunderwood]

Originally Posted by bur

You mean by following your instructions for building ecpp? I did that, but it's not up-to-date, I built it Jan 10. So I'd better build an updated version prior to testing.

You can update all the depending packages and also install GWNUM per the instruction in this post. I am compiling with #define GW_THRESHOLD 2048 and with the three printf statements commented out

Yes, what rights does that user need? Is access to a folder containing the ecpp-mpi binary sufficient?

I need only a directory callled "ecpp" in my home named "paul". I will be calling from your system only ecpp-mpi -- you do not need to copy it.

The system is setup to accept authentication via SSH keys only, will that work? I could change it, but would strongly prefer not to.

I will need a key from you in that case.

I kept these general question here, because it might be interesting to others as well. Also I will need a while to set everything up, building ecpp with the various patches took some time. I think some time next week. Maybe in the meanwhile others will join as well.

That'll be great of you and whoever. We can share the top20 points at t5k.org.