Skip to main content

Spot Instance

A Spot Instance is an Amazon EC2 instance that utilizes unused EC2 capacity, available at a significantly lower cost compared to On-Demand pricing. By leveraging spare capacity, Spot Instances allow you to save on your EC2 expenses. The price for a Spot Instance is referred to as the Spot price, which is determined by Amazon EC2 and fluctuates based on the supply and demand for Spot Instances in each Availability Zone.

Spot Instances are ideal for applications that can tolerate interruptions and flexibility in scheduling. For instance, they are well-suited for use cases like data analysis, batch processing, background tasks, and non-critical workloads. If your applications can adapt to potential interruptions and do not require constant availability, Spot Instances provide an efficient and cost-effective solution.

Workflow for Spot Instance Request and Deployment​

  • The customer navigates to the Nodes section in the TIR platform.
  • Select the Spot Instances tab under Node, where they can view available GPU plans (e.g., GPU type, RAM, price).
  • The TIR platform checks the availability of GPUs based on the request criteria.
  • The system verifies whether the required GPU type is available.
  • If available, the system proceeds to deploy the spot instance immediately.
  • If the required GPUs are not available, the platform moves to the next step.

Steps to create Spot Instance​

  • To create a spot instance you have to click on spot instance under Node.

Click spot instance

  • Select image and click on Next button.

Click spot next

  • Select spot plans and click on Next button.

Select spot plans

  • Click on proceed button to launch spot instance.

Select spot plans

Note

The request is currently unavailable and will be queued for deployment. The estimated deployment time is between 48 to 72 hours. Please note that after deployment, the instance may be shut down by us in the future to reclaim resources as needed. You will be notified via email if this occurs.

GPU Unavailability Notification​

If the requested GPU is not available, the customer is notified via an on-screen message: Message: "The requested GPU type is currently unavailable. Your request will be queued for deployment. Estimated time for deployment: 48-72 hours."A threshold timer (48-72 hours) is set for the customer’s request. The system sets a priority queue for deployment within this time window.

Deployment Within Threshold​

If GPUs become available within the 48-72 hour window, the request is processed, and the customer is notified about the deployment. Notification over email: "Your requested GPU instance is now available and has been deployed successfully." GPU details will be specified to the customer include, the count of deployed GPU, RAM and will be informed that this can be removed at any point of time.

  • The instance is deployed and the customer can use it as per their request.
  • The Spot Instance section under Nodes will update to reflect the deployed instance.

Threshold Expiry (After 72 Hours)​

If the GPU is not available within the 48-72 hour threshold, the customer’s request is automatically expired. Notification: "Your request for a GPU instance has expired due to unavailability beyond the 72-hour threshold. Please submit a new request." The expired request is moved to a closed status in the system, and the customer’s request is no longer considered valid. Re-request or Cancellation The customer has the option to re-submit the request or cancel the request entirely after expiration. Re-submit: The customer may update their GPU requirements and submit a new request. Cancel: The customer may cancel the request, and no further actions will be taken.

State of Spot Instance​

The status could be one of the following:

  • Queuing: Waiting for GPU availability.

While in the queuing state, you can only perform the delete action. Alternatively, you can wait for resource availability.

Action Queuing State

  • Deploying: GPU requested is now available and being deployed.

  • Expired: After 72 hours, the request will expire as the threshold has been exceeded, causing the Spot Instance to transition to the expired state.

Spot Experied

You can relaunch and delete the spot instance on expeired state.

Spot Actions Experied

Relaunch spot instance

  • Running: When the customer's GPU is successfully deployed it is coming on running state.