0

I am working on a deep learning project recently. Because I need a large memory size and GPUs, I tried to create a deep learning VM instance on Google Cloud Platform (GCP). Deep Learning VM is a market place solution for deep learning application. It has a high-memory CPU, a GPU and PyTorch or TensorFlow ready in the VM. However, I usually get this error recently.

ResourceType":"compute.v1.instance","ResourceErrorCode":"ZONE_RESOURCE_POOL_EXHAUSTED","ResourceErrorMessage":"The zone 'projects/flawless-psyche-366003/zones/us-east1-d' does not have enough resources available to fulfill the request. Try a different zone, or try again later."

This indicates that no resource is available in this time zone/ region.

The next thing I did was blindly trying different time zones or regions until I found the right one. I search whether there is a way to find the right time zone efficiently, but not much discussion is related.

Is there any other efficient way to check the resource availability in different time zones/regions on GCP? Then, I can create my VM in the time zone which has more resources without trying every zone randomly.

Thank you!

Mayur Kamble
  • 180
  • 4
Jean Lin
  • 3
  • 1

1 Answers1

2

the GPU resource are rare on the cloud (and beyond). The resources are often exhausted. You can try to run your process when there are less people connected (the evening, the night, the weekend,...);

In any case, there is no API or service to check the availability.

The Google Cloud (expensive) solution, is to create a reservation (and therefore a commitment) of the resource. Like that, you will pay it for 1 or 3 years (with a discount) and you will be sure that this resource is reserved to you.

It's a convenient (and expensive) solution, that make sense if you have a sustainable need of those reserved resources, but a waste of money if you have only sporadic needs.

guillaume blaquiere
  • 66,369
  • 2
  • 47
  • 76