Platform: GCP ✅

Part 1 (2019)
63

It worked for me on Windows 7!

In summary, what I did differently from the tutorial is the following:

  • I replaced “step 2: Install Google CLI from the tutorial” with steps 1, 2 and 3 of the “Before you begin” section of Quickstart for Windows
  • Once the installation was done, in the Google Cloud SDK Shell I then picked my cloud project and then my default region (europe-west1-b in my case) as requested
  • To create an instance (step 3 of the tutorial), I simply typed the following command in the Google Cloud SDK Shell:

gcloud compute instances create “my-fastai-instance” --zone=“europe-west1-b” --image-family=“pytorch-1-0-cu92-experimental” --image-project=deeplearning-platform-release --maintenance-policy=TERMINATE --accelerator=“type=nvidia-tesla-p100,count=1” --machine-type=“n1-standard-8” --boot-disk-size=120GB --metadata=“install-nvidia-driver=True” --preemptible

Then I just followed the rest of the tutorial.

6 Likes
64

I’m really surprised how setup went smoothly. It took me around 10 minutes. I’ve used the guide completly (only altering location to EU).

3 Likes
65

Setup was real smooth for me too on Windows 10. Thanks for the excellent guide. Followed it exactly as it was it written.

2 Likes
66

Hmm I tried using attach-disk while both instances were stopped:

gcloud compute instances attach-disk "fastai-cpu" --disk "fastai-p100-gpu" --zone us-west1-b

However, this is the error that is given:

ERROR: (gcloud.compute.instances.attach-disk) Could not fetch resource:

  • The disk resource projects/.../disks/fastai-p100-gpu is already being used by projects/.../instances/fastai-p100-gpu

Am I missing something?

67

Hi @angert, actually i just updated my reply, please read it. The previous one was not complete. The error is because you didnt detach the disk. Also are you sure your external disk’s name is “fastai-p100-gpu” ? I think its the name of boot disk of your gpu-instance.

68

For a VA/DC resident, would it be ok to change the region/zone to us-east1-c? I originally was going to do us-east4, but didn’t seem to have GPU features, so decided the South Carolina region would be better. Or do you recommend us-west1-b even for us East Coasters?

Also, it never gave me the option to choose zone/region via command line for some reason…

Thanks!

69

When you do gcloud init, then you need to choose option 1[modify the default configuration]. then choose your account, followed by project. after that it asks for default region and zone.

1 Like
70

Makes sense, thanks! As for the region/zone, my understanding is that you want to pick a region/zone close to you for latency purposes. Is there something us-west1-b offers that us-east1-c doesn’t that I should opt for that data center, despite it being farther away?

71

just check if P100 and k80s are available in that region. sometimes pricing also differs!

1 Like
72

Is there terminal commands for starting and stopping the instances?

2 Likes
73

Thanks Jeremy :slight_smile:
I’m using Ubuntu 18.04 on Windows 10 for setting up the GCP environment, and I think that the copying and pasting does not properly work between by my Chrome browser and my Ubuntu on Windows10. I noticed that some extra characters appear when I paste lines of code which are too long to fill in one single line :face_with_raised_eyebrow:
I created a whole new instance and typed manually everything following the instructions, and I did not encounter any issue this time.

74

This seems to work for me, but I’m new to gcloud so let me know if anyone knows of a better way to achieve this:

Starting instance: gcloud compute instances start INSTANCE_NAME
Stopping instance: gcloud compute instances stop INSTANCE_NAME

Replace INSTANCE_NAME with your specific name, which in my case was my-fastai-instance as used in the instructions.

Hope this helps :slight_smile:

6 Likes
75

Start - gcloud compute instances start INSTANCE_NAME
Stop - gcloud compute instances stop INSTANCE_NAME
Optionally you can use --async to just start the operation and not get stuck until it finishes.
Some more cool stuff like GSTART instance_name and other functions i created.
link - Link
Feel free to use them.

2 Likes
76

When I run this command with correct ZONE and INSTANCE_Name, I got an error “(gcloud.compute.ssh) [/usr/bin/ssh] exited with return code [255].”

Do you have any idea? Thank you.

1 Like
77

Try gcloud compute ssh instance-name -vvv. This will show the exact problem if there’s any.

2 Likes
78

I got the error says:
(gcloud.compute.ssh) unrecognized arguments: -vvv

79

Hi all,

Hoping for some guidance with setting up GCP with the Mac Terminal.

I have successfully completed steps 1 and 2 of the setup guide and I received the following terminal output after changing my zone to Australia

Your Google Cloud SDK is configured and ready to use!

* Commands that require authentication will use my_email@gmail.com by default
* Commands will reference project `fastai-v3-220XXX` by default
* Compute Engine commands will use region `australia-southeast1` by default
* Compute Engine commands will use zone `australia-southeast1-b` by default

However, when I run the Step 3 code in my terminal, nothing happens. I have waited over 10 mins, so I presume I am doing something wrong. Please see the below where I am attempting to complete step 3 with no joy.

34%20am
Screenshot 2018-10-24 at 9.24.34 am.jpg1228×358 99.6 KB

I have some experience with the terminal, but by far no master of this space.

I have checked that account is correctly set, with the following command and output

Screenshot%20at%20Oct%2024%209-32-45%20am
Screenshot at Oct 24 9-32-45 am.jpg712×212 37.5 KB

Thanks in advance for any help anyone is able to provide.

Todd

1 Like
80

having the following character:
“>” means that the command NOT YET executed

Probably there are some problems with " or ’ , try to copy paste command to editor, replace all " and ’ with " and try again.

3 Likes
81

Hi, thanks for your help. That was all that was causing me the problem. All setup and running now. A good trick to remember! Many thanks. Todd

1 Like
82

There’s a new GCP image just available. If you use it to create a brand new instance, then you no longer have to ‘git clone’. The version of the course in your folder is ready to go automatically, and you can simply cd to it and ‘git pull’ to get the latest version of the course repo.

4 Likes