You can use the Google Cloud Pricing Calculator to total the estimated monthly costs for each GCP product.
You dont incur any charges for doing so.
Ref: https://cloud.google.com/products/calculator

A managed instance group (MIG) is a group of identical virtual machines (VMs) that you can manage as a single entity. You can use a MIG to deploy and maintain a stateless application that runs directly on VMs. A MIG can automatically scale the number of VMs based on the load or a schedule. A MIG can also automatically heal the VMs if they become unhealthy or unavailable. A MIG is suitable for applications that need to run on VMs rather than containers or serverless platforms.
B is incorrect because Kubernetes Engine is a managed service for running containerized applications on a cluster of nodes. It is not necessary to use Kubernetes Engine if the application does not use containers and can run directly on VMs.
C is incorrect because Cloud Functions is a serverless platform for running event-driven code in response to triggers. It is not suitable for applications that need to run continuously and handle HTTP requests.
D is incorrect because Cloud Run is a serverless platform for running stateless containerized applications. It is not suitable for applications that do not use containers and can run directly on VMs.
Managed instance groups documentation
Choosing a compute option for Google Cloud

The Vertical Pod Autoscaler (VPA) in Kubernetes automatically adjusts the CPU and memory requests and limits of the containers within a pod based on historical and real-time resource usage. In this scenario, where a single-replica stateful application needs more CPU during peak times, VPA can dynamically increase the CPU allocated to the pod when needed and potentially decrease it during off-peak periods to optimize resource utilization and cost efficiency.
Option A: Cluster autoscaling adds or removes nodes in your GKE cluster based on the resource requests of your pods. While it can help with overall cluster capacity, it oesn't directly address the need for more CPU for a specific pod.
Option C: Horizontal Pod Autoscaler (HPA) scales the number of pod replicas based on observed CPU utilization or other select metrics. Since the application can only have one replica, HPA is not suitable.
Option D: Node auto-provisioning is similar to cluster autoscaling, automatically creating and deleting node pools based on workload demands. It doesn't directly manage the resources of individual pods.
Reference to Google Cloud Certified - Associate Cloud Engineer Documents:
The functionality and use cases of the Vertical Pod Autoscaler (VPA) are detailed in the Google Kubernetes Engine documentation, specifically within the resource management and autoscaling sections. Understanding how VPA can dynamically adjust pod resources is relevant to the Associate Cloud Engineer certification.

Let's evaluate each option based on the requirement of sub-millisecond latency for globally stored IoT data:
A: Spanner with Caching: While Spanner offers strong consistency and global scalability, the base latency might not consistently be sub-millisecond for all read/write operations globally. Introducing caching adds complexity and doesn't guarantee sub-millisecond latency for all initial reads or cache misses.
B: Bigtable: Bigtable is a highly scalable NoSQL database service designed for low-latency, high-throughput workloads. It excels at storing and retrieving large volumes of time-series data, which is typical for IoT sensor data. Its architecture is optimized for single-key lookups and scans, providing consistent sub-millisecond latency, making it a strong candidate for this use case.
C: BigQuery: BigQuery is a fully managed, serverless data warehouse designed for analytical queries on large datasets. While it's excellent for analyzing IoT data in batch, it's not optimized for the low-latency, high- throughput ingestion and retrieval required for real-time IoT applications with sub-millisecond latency needs.
D: Cloud Storage with Cloud CDN: Cloud Storage is object storage and is not designed for low-latency transactional workloads. Cloud CDN is a content delivery network that caches content closer to users for faster delivery, but it's not suitable for the primary storage of rapidly incoming IoT sensor data requiring sub- millisecond write latency.
Google Cloud Documentation References:
Cloud Bigtable Overview: https://cloud.google.com/bigtable/docs/overview - This document highlights Bigtable's suitability for low-latency and high-throughput applications, including IoT. It mentions its ability to handle massive amounts of data with consistent performance.
Spanner Overview: https://cloud.google.com/spanner/docs/overview - While Spanner offers low latency, Bigtable is generally preferred for extremely high-throughput, low-latency use cases like raw sensor data ingestion due to its optimized architecture for such workloads.
BigQuery Overview: https://cloud.google.com/bigquery/docs/introduction - This emphasizes BigQuery's analytical capabilities rather than low-latency operational workloads.
Cloud Storage Overview: https://cloud.google.com/storage/docs/overview - This describes Cloud Storage as object storage, not ideal for sub-millisecond latency reads and writes required for real-time IoT data.

Google Cloud provides Cloud Audit Logs, which is an integral part of Cloud Logging. It consists of two log streams for each project: Admin Activity and Data Access, which are generated by Google Cloud services to help you answer the question of who did what, where, and when? within your Google Cloud projects.
Ref: https://cloud.google.com/iam/docs/job-functions/auditing#scenario_external_auditors