[QUOTE=bayanne;527648]What happens when you reach the 30 hours runtime within a week?
Should one just stop, or will Google terminate your connection, or what?[/QUOTE]
They will terminate your instance after a while, I guess it depends on free resources and requests they have, they will let you live longer, for me it was like ~32, 46, 34 hours in the last three consecutive weeks, in chronological order. No joke, we got over 45 hours on our second week, and over 33 hours this week.

On the other hand, on colab, we didn't get a T4 anymore, regardless of how many time we reconnect and deleted the notebooks, I guess they learned we don't play orthodox. Next step is to see how do we do to use the K80 for cudaLucas, where it would be more efficient, this card is a waste to be used for TF (well, it is for free... we should not complain!)

[QUOTE=chalsall;527659]LOL... Thanks, guys.

You know you're doing an OK job when the bug reports you're getting are smelling mistakes in the Human... :wink:[/QUOTE]The spell check seems to be hard wired in my head. I've not provided any bug reports, because I've not yet attempted to run Colaboratory or any of the related scripts. So there's currently no data here showing whether the code error rate is similar to the spelling error rate. When the easily checked and easily corrected visible things have a high error rate it leaves me wondering about the rest.
Writing is hard. Being a critic is less so. All productive effort is appreciated.

[QUOTE=kriesel;527663]When the easily checked and easily corrected visible things have a high error rate it leaves me wondering about the rest. Writing is hard. Being a critic is less so. All productive effort is appreciated.[/QUOTE]

I copy you. And I very much appreciate all criticism. It's how things get better.

But to put on the table for a bit of context... Many of the pages I build for Humans I don't actually look at myself all that often.

Secondly, I'm ***seriously*** dyslexic. I actually almost dropped out of high school because I couldn't spell. I was saved by a wonderful English teacher who took me aside one day and said "Don't worry about the spelling, Chris. Just write."

It's one of the reasons I have such a passion for education. A good teacher can change a person's life! :tu:

Google Colab just keeps disconnecting after running it for 1 hour. It consistently happen everytime a new instance is created and compute starts, and it has been going on for the past week or so. But if I click reconnect within 15 minutes after it says disconnected, I can keep running it smoothly for the next 12 hours. Any ways of bypassing that?

[QUOTE=xx005fs;527705]Google Colab just keeps disconnecting after running it for 1 hour. It consistently happen everytime a new instance is created and compute starts, and it has been going on for the past week or so. But if I click reconnect within 15 minutes after it says disconnected, I can keep running it smoothly for the next 12 hours. Any ways of bypassing that?[/QUOTE]

I had a similar problem once, and concluded I was actually resuming my previous session instead of getting a new session. I have found that if I make sure I have a fresh session by using Runtime --> Manage Sessions --> Terminate Session, then connecting again (it makes you get a new authentication for Google Drive), I can always get the full 12 hours.

I've been successfully using Colab this week, but as of this AM I'm unable to get a GPU backend. Is this throttling or just random unavailability? Any way to tell? If throttling, is there a timeframe in which you can once again get a GPU?

[QUOTE=mnd9;527732]Is this throttling or just random unavailability? Any way to tell? If throttling, is there a timeframe in which you can once again get a GPU?[/QUOTE]

It's impossible to tell.

Right now, for example, from my main workstation I have three instances running (started at around 1200 UTC). One under my main Google account, and two under a second (different tabs, same browser).

Over on my RPi test workstation, on the other hand, I can't get an instance, even though that is SOCKS tunneled to appear to be in North America.

I've found it's best to just try every couple of hours if you don't first succeed. But don't get too pushy -- Colab can get "grumpy", and will occasionally give you an instance only to ungracefully kill it a few minutes later...

[QUOTE=chalsall;527735]It's impossible to tell.

Right now, for example, from my main workstation I have three instances running (started at around 1200 UTC). One under my main Google account, and two under a second (different tabs, same browser).

Over on my RPi test workstation, on the other hand, I can't get an instance, even though that is SOCKS tunneled to appear to be in North America.

I've found it's best to just try every couple of hours if you don't first succeed. But don't get too pushy -- Colab can get "grumpy", and will occasionally give you an instance only to ungracefully kill it a few minutes later...[/QUOTE]

That's exactly what just happened--got a GPU instance, only to have it killed after 5 min!

Use alternating signons
 

I created 2 different Google accounts and I alternate between them. After the 12 hour disconnect, I switch to the other one etc. So far that has eliminated the disconnects for me.

[QUOTE=Chuck;527744]So far that has eliminated the disconnects for me.[/QUOTE]

Clever! :tu:

Today I'm running an experiment wrt possible daily quotas per account... One account is running one instance, the other two. It will be interesting to see if the latter gets killed after six hours. And then, what the relaunch success rates are.

I still don't understand why my RPi can't get an instance -- as far as Google should be able to tell, I'm sitting in front of a console in a machine room somewhere in the eastern USA. And it /did/ get an instance about half an hour ago -- died five minutes later...

P.S. Ahh... Maybe Google is smart enough to know that I /shouldn't/ be in a machine room...

So are you guys talking about GPU related disconnects?

I am running two different accounts concurrently, but from the same IP address. Each runs basically 24/7 because I reconnect each 12 hours, as soon they end.

I can connect every time, and as long as I insure that I am getting a new connection, I never seem to get disconnected prematurely,

But I don't need any GPU resources, so maybe that is the difference.