[QUOTE=Uncwilly;530021]Has anyone that has been working on using the CPU's tried to set it up under one of the GPU72 sponsored sessions? If you have and it works, then getting Chris to include that in the deployment could get us some nice P-1 or DC work too.[/QUOTE]

I've mapped a solution space for this. It will be brought in as a secondary payload to the TF'ing Perl script.

I'm also looking at providing a more generic payload mechanism, which will allow choosing the work types for both the GPU and CPU compute available. Please note though that the CPUs available are not, really, all that powerful. There do seem to be a lot of them available, though...

In the short term, what I've been doing during development is to launch a tunnel building Section before the TF'ing section (these can safely run in immediate succession -- as in you can "Commit" them on Kaggle if desired).

I then SSH into the instance, setup whatever CPU job I want to run, and then launch it.

Please don't expect me to have any code implementing any of this out of alpha for at least a week. Lots going on, not just here...

Edit: Oh, and Uncwilly: I just *love* that you're now regularly controlling what would have once been considered a supercomputer by way of a device that you carry around in your pocket or on your belt! :bow:

[QUOTE=chalsall;529959]1. Yup. Tee works too.

1.1. You guys are going to /love/ Linux. You really are! :wink:

2. Really? Hmmm...
[/QUOTE]
1.1 We'll see. Each OS has its pros and cons. Having used DOS & Win since 1982 and 1992, learning curve definitely counts against linux, especially for us older users with porous cranial memory. Perhaps if I'd gotten into linux in 1992. I've tried linux many times over decades, still have slackware and redhat media around from early days. What I really miss is VMS with its superb unmatched documentation and command nomenclature that's clear, not cryptic. But not the painfully slow and costly hardware it ran on. (One of the early $10,000 graphics hardware offerings from DEC was fiber interfaced to one of their minicomputers, and was so slow in combination with my employer's mini, that it was quicker to plot an image with a pen plotter on paper than to get it loaded to the screen!)

2. May have been a leapt conclusion too far. There seems to be about a 10 minute lag between live results and Google Drive update.

[QUOTE=kriesel;530025]What I really miss is VMS with its superb unmatched documentation and command nomenclature that's clear, not cryptic.[/QUOTE]

I will agree with that. The *nix command space is a mish-mash of history. As I said earlier, it annoys me that the "mlocate" package managed to claim "updatedb" is its own.

But this is where we (including you) find ourselves.

In absolutely no way meaning to come across as condescending, but the most often encounter command-line environment is going to be Linux -- for just about all contemporary ICT infrastructure. From your servers to your routers to your firewalls to your TVs to your phones to your IoTs et al.

Linux runs the Internet. Even Micro$oft has finally admitted this.

I would argue that we "old-timers" can all reminisce about how long we've been in the game, and all the amazing things we did to squeeze every last cycle out of a machine. But, perhaps, there would be more "upside" if we were to teach people how to use what is available right now.

To briefly share, some of my other work involves trying to figure out how to get more young people to be "technology creaters" rather than "technology consumers".

A non-trivial problem space.

So many people spend so much of their day staring at a small little device in their hands (sometimes with ear-phones, so they're fully immersed). Sadly, not producing but rather consuming...

[QUOTE=kriesel;530025]2. May have been a leapt conclusion too far. There seems to be about a 10 minute lag between live results and Google Drive update.[/QUOTE]

I was taught by a trusted mentor that this kind of work is best done with a pen and paper pad beside you.

Run experiments. Observe and write down the empirical results (with timestamps).

Repeat as necessary... :smile:

[QUOTE=chalsall;530036]I will agree with that. The *nix command space is a mish-mash of history. As I said earlier, it annoys me that the "mlocate" package managed to claim "updatedb" is its own.

But this is where we (including you) find ourselves.
...

I was taught by a trusted mentor that this kind of work is best done with a pen and paper pad beside you.

Run experiments. Observe and write down the empirical results (with timestamps).

Repeat as necessary... :smile:[/QUOTE]I do like the pricing structure of linux (especially now that downloads are possible again without driving to town for them) FIBER! Also OpenOffice etc.

Run experiments! Write things down?! Have you seen any of [URL]https://www.mersenneforum.org/showthread.php?p=521922#post521922?[/URL] Bug and wish lists, how-tos, run time scaling guides based on experiment, and perpetually kept logs of various applications, etc. Bit by bit, it has grown to book length.

On further reflection I'm sure you have seen some of it, since you posted appreciation for the Colab thread at one point.

[QUOTE=kriesel;530052]On further reflection I'm sure you have seen some of it, since you posted appreciation for the Colab thread at one point.[/QUOTE]

I have seen all of it. Appreciatively. :smile:

Not enough people understand just how important knowledge curation is.

As The Musk itself has said: "People assume that technology just automatically advances. It doesn't. It takes a lot of work to move forward. Entropy is not on your side.

[QUOTE=chalsall;530053]As The Musk itself has said: [/QUOTE]The [U]Venerable[/U] Musk.

Got my account blocked in kaggle. Emailed support and got it unblocked, but...
[QUOTE]Thank you for your inquiry. Your account has been blocked for using mprime to search for prime numbers. This is not allowed as it is over-utilizing our resources.

We have unblocked your account with a warning. [/QUOTE]
For now, I am doing a hard stop on kaggle. YMMV.

[QUOTE=axn;530087]...For now, I am doing a hard stop on kaggle. YMMV.[/QUOTE]

Both of my [I]Colaboratory[/I] instances shut down about 12 hours ago. No GPU and/or bandwidth available, it implies. When it gives me access again, I plan to run only one instance.

[QUOTE=axn;530087]Got my account blocked in kaggle. Emailed support and got it unblocked, but...[/QUOTE]

Hmmm... Because I consider this critical, I am including below the full email I received from Kaggle's support team telling me what we were doing was OK.

To get pedantic, the email did say "GPUs", but since they're so much more coveted than CPUs, I would have thought CPU usage would have been fine as well.

Thoughts?

[CODE]Date: Fri, 11 Oct 2019 21:30:12 +0000
From: Kaggle Support <support@kaggle.com>
Reply-To: support@kaggle.com
To: Chris Halsall <chalsall@[REDACTED]>
Subject: [Kaggle] Re: I have a different problem [#20190909143005]

Your request (28491) has been updated. To add additional comments, reply to=
this email.
----------------------------------------------

Kaggle Support, Oct 11, 2:30 PM PDT

Hello Chris,

Thank you for your patience. The linked thread provided doesn't describe in=
adequate detail what your team intends to do. The discussion is mainly fro=
m a collaboration aspect, and the link is to a private GitHub repo.

However, if your team is trying to identify Mersenne prime numbers by using=
Kaggle GPUs, that is totally fine.


Regards,
Kaggle Support

----------------------------------------------

Kaggle Support, Sep 18, 7:33 PM PDT

Hello Chris,

Thank you for your inquiry. The team is currently discussing and looking in=
to this issue.

We will reply back as soon as we receive an update.

Regards,
Kaggle Support

----------------------------------------------

Chris Halsall, Sep 9, 7:30 AM PDT

User Email: chalsall@[REDACTED] User Display Name: Chris Halsall Question: I=
have a different problem DifferentQuestion: Hello. I'm a member of the Gre=
at Internet Mersenne Prime Search (GIMPS) community -- a group of computer =
and mathematics geeks who work together to find very large Prime numbers. W=
e recently came across Kaggle and Colaboratory, and were quite surprised by=
how easy it is to configure Notebooks to use the free GPU resources provid=
ed to assist in our search. Please see https://mersenneforum.org/showthread=
.php?t=3D24646#post525526...for the discussion and experiments we've been d=
oing. However, we're concerned that this type of work might not be consider=
ed "fair-use" of these systems. We don't want to be doing things which are =
considered inappropriate. So... Could you please let me know what Kaggle's =
position is on this, and I'll share your response with our community. Thank=
s and regards. -Chris[/CODE]

I have a feeling that email was before many here got on the Kaggle bandwagon. But now they realize what has been unleashed and have quickly reversed their stance.

It is also possible that the service agents don't have clear rules on what is allowed and what isn't, so some decisions are left to the discretion (and mood) of the individual agents.

[QUOTE=PhilF;530130]I have a feeling that email was before many here got on the Kaggle bandwagon. But now they realize what has been unleashed and have quickly reversed their stance.[/QUOTE]

I really doubt that we're even a rounding error on their usage graphs.

[QUOTE=PhilF;530130]It is also possible that the service agents don't have clear rules on what is allowed and what isn't, so some decisions are left to the discretion (and mood) of the individual agents.[/QUOTE]

Much more likely. Which is why I thought it important to document the prior exchange with them.

Precedent is an important word... :smile: