Not your average Puppy box

[prehistoric]

@gcmartin

I'm not really shopping for distros for this box. I've tried several, including Ubuntu, Linux Mint, Fatdog and some other Puplets I haven't listed. My problem is that Quirky 7.0.4.1 handles all the installed hardware, while older kernels fail on things like the gigabit network adapter. It is apparent this early board was built with a buggy 8169 chip from Realtek, a company which has created problems for any number of Linux systems. The 3.19 kernel in Quirky has code to work around the problem. I also must have CPU scaling to avoid overheating.

I'm not very worried about user interface. When I run a session with heavy computing I typically drop back to the console interface to cut overhead from system tasks. I did this with those tests computing Pi and gamma to ridiculous numbers of decimal digits.

My comment about compiling was intended to limit requests to build stuff where I have to figure out a complicated build environment and tweak large numbers of flags. If someone else sets things up so I can build stuff while the computer does all the heavy lifting I'm amendable to that approach.

I'm fairly certain I do not want to run a server. I have enough trouble maintaining a few systems for people I know well. The Internet is a jungle, and defense against predators becomes a full-time job.

What I have been thinking about are combinatorial optimization problems for which I can get programs already designed. I would like problems for this to be as independent of computing environment as possible. This gets me out of the loop of constantly fighting with computing environments which are convenient for somebody else.

An extreme example would be problems and solutions in the form of tables or graph structures which could be transferred as comma-separated lists, with a simple description of the criteria to measure value of a proposed solution. I think I could encode such descriptions in formulas to be passed to optimization programs.

This is all a kind of thought experiment to drill down to the question of what can you use computational power for, other than running more and more complicated systems which themselves consume not only computer memory and processor cycles, but also user time and mental effort.

[gcmartin]

... use computational power for ...

Validation. Simulation. Mathematical description of Physical/Chemical/Biological phenomenon. Predictor of Quantum Behavior. Calculation of the date of the Sun's collapse which will consume Earth in it collapse (a primary reasoning behind the Mars colonization). Seti.

All kinds of efforts that "computationals" help mankind with.

[prehistoric]

... use computational power for ...

Validation. Simulation. Mathematical description of Physical/Chemical/Biological phenomenon. Predictor of Quantum Behavior. Calculation of the date of the Sun's collapse which will consume Earth in it collapse (a primary reasoning behind the Mars colonization). Seti.

All kinds of efforts that "computationals" help mankind with.

I'll admit that the possibility the Sun will consume the Earth in 500 million years does not panic me. Even there, I'm satisfied with modest calculations already done.

Surprisingly, mathematical models of stellar interiors seem to be pretty good in the region where radiative transfer of energy predominates. The modeling problems set in about half way to the surface, along with convective transport, which moves stuff toward the visible surface at about walking speed. Moving plasma also generates magnetic fields, and by the time these reach the corona they can dominate gravitation forces in some prominences. If we really understood magnetohydrodynamics, we'd have working fusion power generators. Last time I checked we were still short of modeling the full complexities of tokamaks.

Modeling the convective region is very important to understanding mixing, and mixing is a major factor in composition of stellar interiors. This is a primary variable in the evolution of main-sequence stars like the Sun.

If all that is not enough, there is a problem of mass loss from the surface via "solar wind". This will pick up as the Sun ages. We can expect the Sun to lose a substantial fraction of its total mass before it becomes a red giant. This is another big factor in those predictions.

Fortunately, we, or the species which replaces us, will have hundreds of millions of years to figure things out.

Could we have some more suggestions appropriate for this century?

[prehistoric]

Here's a worthy cause concerned with discovering how proteins fold, which is an important aspect of many diseases: Folding@Home.

This can run on some Puppy derivatives related to Ubuntu, but unfortunately I have not been able to get my Ubuntu versions to correctly handle the buggy RTL8169 gigabit Ethernet interface on this box. Internet connectivity is necessary for this project. I will definitely want to run a 64-bit version.

Getting this to run on Quirky 7 would allow me to keep the computer busier than its owner.

Added: I've now taken wyzguy's advice, and changed the IOMMU setting in the BIOS. Also changed some related IO settings. There is no particular problem with that RTL8169 on common 64-bit distros.

[Deacon]

So far, this machine has demonstrated stability by calculating Pi to 5,000,000,000 decimal places in a little over half an hour. I've still got more to do in convincing existing computational software it is not a good idea to put huge temporary files in the same place as the OS.

I'm open to suggestions for what you do with this much computational power, besides running servers or playing games. (My only reason for that rebuilt Radeon card is to get video memory bandwidth off the processor-memory bus.) This is a machine one could scarcely have imagined when I started in computing. The millennium has arrived. What do we do next?

Mine cryptocurrencies and donate them to "Deacon's Hamburger Fund". (I will gladly pay you Tuesday.) I also accept PayPal.

[Colonel Panic]

Out of interest prehistoric, what computational software do you use to calculate all those decimal places of pi? I've managed about 850,000 with a DOS-based program but not got anywhere near 5,000,000,000.

[prehistoric]

I used y-cruncher 0.6.8 build 9461. The particular algorithm was the Chudnovsky one based on their series that yields 14 digits per term. This by itself would not scale well, but it was then transformed into a binary-splitting algorithm for summing that series.

The same program will also compute such constants as Euler's gamma.

The interface is clunky, but I'm not looking for eye-candy. I run it from a pure console environment.

[Colonel Panic]

I used y-cruncher 0.6.8 build 9461. The particular algorithm was the Chudnovsky one based on their series that yields 14 digits per term. This by itself would not scale well, but it was then transformed into a binary-splitting algorithm for summing that series.

The same program will also compute such constants as Euler's gamma.

The interface is clunky, but I'm not looking for eye-candy. I run it from a pure console environment.

Thanks, I'm downloading it now.

[Lobster]

It sounds like the ideal machine for ethical penetration testing, in particular brute force password cracking.

You could also use it as a morphing firewall, running virtual box honeypots, different random layers etc.

In terms of software, 3d, compiling as mentioned, neural processing and I rather like the idea of running as many different OS as possible but keeping the data on a Puppy multi session. Then trying to access the multisession in various ways ...

Just ideas ...

[prehistoric]

Hi Lobster,

I was beginning to wonder what had happened to you.

I always tend to worry about what has happened to lobsters I know when we get to the part of the calendar the French revolution named Thermidor.

I'm not very interested in penetration testing, because there is abundant evidence people with far more resources are already busy doing more than I ever could.

I've been wanting to find a combinatorial optimization problem related to Puppy which would allow us to pack an "impossible" amount of software into a small system. Gifted individuals have already done a great deal in this direction, but there remain design tasks that overwhelm them, leading to experiments in system building that take a lot of time and effort.

Can you think of design problems which might be abstracted enough to be tackled this way by someone who depends on others to remember all the idiosyncratic details of system building?

[Lobster]

Good to hear we lobsters have our own revolution.

I am not sure if this answers your question but quantum storage will become a norm in the future. That allows massive in fact impossible storage.

The sort of hardware you have might be used for quantum and qbit programming simulation - there are languages, some open source.

The other people using impossible data compression are NASA, the problem is there decrypting is done (of packets of info) with supercomputers. Some of the decrypting of old generation software might be open sourced. The NSA might have similar software some of which is available on their site for free, some for world use, some for uSa residents only . . .

[gcmartin]

URL lost. Post erased.