ideas for a cloud based pet project?

[ET_]

With lots of RAM you could do some deep ECM on GIMPS and Fermat numbers as well.

Or use the mtsieve framework to help sieving (and testing) projects like FermatSearch and DoubleMersennes.

Or use a swarm of processors to ECM process numbers in the Brent tables using ECMNET.

[xilman]

Originally Posted by ET_

Or use a swarm of processors to ECM process numbers in the Brent tables using ECMNET.

Or generalized Cullen & Woodall numbers or homogeneous Cunninghams for me and Jon Becker respectively.

[kriesel]

Originally Posted by Prime95

Well if every atom in the universe can hold one bit....

How many universes are there in the multiverse? I ran some numbers a while ago for a more ambitious thought experiment, storing 10 data bits per proton mass in the observable universe (including the estimated dark matter), in some combination of position, spin, and momentum. That data storage capacity allows for about 263 iterations, with some other assumptions mixed in. But the clock rate would become really slow, given that most of the mass is scattered over millions if not billions of light-years. And I for one would strenuously object to having my hydrogen content repurposed for data storage while I'm still using it for biochemistry.

https://www.mersenneforum.org/showpo...97&postcount=6

[chris2be8]

Originally Posted by ET_

Or use a swarm of processors to ECM process numbers in the Brent tables using ECMNET.

If I read https://www.slideserve.com/brant/fro...-brent-factors correctly then all of the Brent tables have been factored to T60. So that's probably not worth doing, unless you plan to extend the brent tables (I've found a few easy ecm factors just above the 10^255 limit).

Chris

[xilman]

Originally Posted by Prime95

Well if every atom in the universe can hold one bit....

Some while back I observed that in principle a single hydrogen atom can store an arbitrarily large number of bits. Implementation is likely to be a real bugger (which is an English colloquialism meaning "non-trivial").

The secret is to store a number as the principle quantum number in a Rydberg atom.

[aurashift]

Keep in mind this should also be storage and maybe networking intensive as well as CPU/RAM-wise, as that's a big aspect of this redesign I'm doing. The fact that ycruncher is writing 5-10 PB is wonderful for this, otherwise I'd have to simulate that kind of workload and get nothing out of it besides some figures that look nice in a powerpoint.

[xilman]

Originally Posted by aurashift

Keep in mind this should also be storage and maybe networking intensive as well as CPU/RAM-wise, as that's a big aspect of this redesign I'm doing. The fact that ycruncher is writing 5-10 PB is wonderful for this, otherwise I'd have to simulate that kind of workload and get nothing out of it besides some figures that look nice in a powerpoint.

In that case, consider running the CADO-NFS siever. You can use as many petabytes as you wish to store the relations. As long as you run many relatively small factorizations there are people around here who can perform the subsequent matrix and square root phases. For the sort of numbers in which I am presently interested, each factorization takes up around 20-100GB which can be downloaded over my internet connection relatively easily. I'd have to think about 1TB data sets.

[aurashift]

Originally Posted by xilman

In that case, consider running the CADO-NFS siever. You can use as many petabytes as you wish to store the relations. As long as you run many relatively small factorizations there are people around here who can perform the subsequent matrix and square root phases. For the sort of numbers in which I am presently interested, each factorization takes up around 20-100GB which can be downloaded over my internet connection relatively easily. I'd have to think about 1TB data sets.

Depending on the quantity and capacity of drives we get, lets theoretically just say 500TB of raw space, 5PBs perfectly and evenly written would only represent two days worth of rated workload. Typically enterprise drives are rated for 5 drive writes per day.

[xilman]

Originally Posted by aurashift

Depending on the quantity and capacity of drives we get, lets theoretically just say 500TB of raw space, 5PBs perfectly and evenly written would only represent two days worth of rated workload. Typically enterprise drives are rated for 5 drive writes per day.

Well, if you want to raise the rate artificially to an arbitrarily large extent, just copy the files back and forth as many times as you wish. The computed data is still useful in some sense.

[a1call]

Assuming you can pool all the available drives, you could record the world's largest primorial ever calculated.
https://www.mersenneforum.org/showpo...0&postcount=35
It will read and write the whole set of drives constantly during the operation.
But it won't be quite as useless as the world record pi calculation though.

[hansl]

Originally Posted by a1call

But it won't be quite as useless as the world record pi calculation though.

How so?