With Compute Engine, Google manages all hardware. You don't need to provision or pre-provision your computer engine. You don't have to worry about where your Compute Engine is, whether it's in Frankfurt or New York since it solely depends on your demand. It is quite elastic. For example, ten years ago we used the on-premise Compute Engine and we constantly had to worry about scalability. With Google Cloud Compute Engine, there is no need to worry about capacity. If you need it bigger, you can upgrade it. If you need it smaller, you can downgrade. Another thing I like about it is that you pay-as-you-go. I think that the minimum you have to pay for Compute Engine is only one hour. If you don't have to use it, you can just stop it and save a lot of money that way. If you wish to use it continuously, Google offers you a discounted contract. It is very cost-effective and I like its availability as well. 

About eighty people at our company use Google Compute Engine. The data engineering team and the data science team use it.

One of GCE's best features is the managed instance groups. We typically use managed instance groups for high availability. You can set certain parameters for managed instance groups where if the load of the computer or server increases beyond 80%, for example, the solution will automatically spawn another instance, and the load will be automatically divided between two systems. If the load is 80% of one of the VMs or GCEs, once the load is divided, it comes down to 40%, so the availability of your systems goes up. However, that all depends on the parameters or configurations we put on the instance group.

You also have regular health checks on these managed instance groups, which are configurable. If these health checks determine something wrong with the VM, they will automatically kick off or spawn a new GCE instance. This way, the outage time is less. Previously, on-premises, unless somebody reported the issue to the helpdesk saying that a particular service was unavailable, then a support team would need to troubleshoot what went wrong, which takes a long time. At least 30 minutes to one hour. But by using these managed instance groups, we can reduce the outage time, and second, we can configure them with minimal resources, bringing down our cost. And if the load increases, the managed instance groups automatically respond to new things. Subsequently, our costs decrease.

We have a wide range of VMs. There are general-purpose VMs that can be used for hosting general-purpose applications. If some of our applications are memory intensive, then we have a lot of VMs in the M1 series. We can use a range of memory-optimized VMs for these things. We have C-series VMs for compute-intensive applications. If we use some mathematical formulas and require a very high throughput from that, there are GPU-optimized VMs used for machine learning or 3D visualizations in rendering software. GPU-enabled VMs are pretty powerful and responsive.

Again, the best part is that we can spin them up when we need them, and once we're done with our work, we can shut them down, allowing tremendous cost savings for any customer. Previously, if we wanted a very high-configuration VM, we had to own the entire hardware and have it on our on-prem data center. And once we'd done with a particular activity, the system would just be lying there on our premises. That is not the case now. We use and decommission it, so we're only billed for the time we're using the product. One of the best things is the preemptible VMs or Spot VMs. These are the cheapest VMs in Google Cloud, but it has a string attached to it where Google can shut down these VMs whenever Google teams split. You only get about 90 seconds notice before they shut down this particular VM. There are scenarios where customers can use these preemptible VMs, for example, when running a batch job. Batch jobs are run once or twice daily, depending on the customer's requirement. Once we are done running these batches, we can decommission the VM. Even if, in the middle of this batch job, Google shuts down these VMs, we can pick up the processing from wherever the VM left off. These are some of the beautiful things we have on Google Cloud concerning the Compute Engine.

It's the managed instance. You never feel any delay or service. If something happens, it will automatically create a new instance with the same configuration.  

The solution is readily available, and software engineers can provision it. It is scalable and allows self-service. 

From a feature perspective, I find API integration, automation capabilities, and features like preemptive and Spot instances valuable. Migration tools have also been useful.  

I also value the ease of adoption and deployment for my engineering teams. The speed at which I can bring up instances is crucial. Migrating from on-premises to Compute Engine was seamless. Additionally, Google's storage layer, including persistent disks and local SSDs, provides better throughput compared to other providers. The availability of GPUs and TPUs is another advantage.

On the console itself, there is something called the recommendations that it provides. So, if I'm underutilizing or overutilizing a machine, it gives me insight into whether I should under-provision my machine or switch to a lower config machine to save some cost. So, that recommendation feature is quite useful.  So, overall, I like the solution's automation, migration, and API integration.

The solution helps to direct SSH into the machine at the click of a button. It also helps to deploy container images right from the UI. There is no need to manage the containers on the machine. I also like the tool’s Spot provision model.

The most valuable feature is auto-scaling. 

The support for ephemeral instances has been particularly valuable for me. It allows me to significantly reduce costs for temporary virtual machines by automatically destroying them once they are no longer needed, which can result in cost savings of up to 90 percent. Additionally, the solution is easy to use.

The solution is highly available, and you can choose your security, such as creating firewall rules. 

One of the most valuable features that Cloud Datalab has is that it integrates seamlessly with other Google Cloud Platform products. You can use Google Cloud Datalab to explore, analyze, transform and visualize data and build machine learning models using existing data in Google Cloud Storage or BigQuery. Cloud Datalab is interactive, so you can run portions of your code and see the results immediately as you work through your Datalab notebook.

Documentation is the best feature of Google Compute Engine. For example, the other day, I wanted to get something running with Nested Virtualization. So, I was able to do that just by going through Google's documentation. That's pretty good.

All of the cloud solution providers are basically the same. There's a little bit around the edges that is different. Each provider has its own kind of unique products but for the most part, 90% of the coverage between all of them is the same.

Compared to other IaaS providers, I believe that Google Compute Engine (GCE) stands out with the ability to customize the virtual machines. With other providers, these are pre-determined specifications. With this you can create machines specific to your application.