Cost of deploying many smaller programs via GCP [closed]

Question

Our company has many customers whose hardware we support. I'd like to create a series of anomaly detection programs (ADs), e.g. using TensorFlow, to monitor their accounts for unusual disk activity, spikes in latency, etc.

I'm hoping to use GCP to help solve the problem and host the programs that we design, but first I've been asked to give an estimate of how much it will cost. Management is worried that creating and deploying hundreds of these ADs will take lots of disk space and involve significant ongoing cost.

I am but a humble data scientist, plus relatively new to GCP, and therefore don't feel that I'm in a position to give a definitive answer, even though my intuition tells me that this is what GCP was designed for and should be able to handle it without breaking the bank.

Can anyone more knowledgable in this area either back me up and/or warn me about the logistics of deploying many, many smaller programs via GCP?

Answer 1

This is more of a capacity/resource utilization planning which does not have a direct answer. Since you are new to GCP, this will solemnly depend on your instance configuration, zone, utilization, monitoring tools, etc. Unfortunately, Google may not be able to provide you with the exact cost of deploying apps within the platform. However, you can start your planning by selecting a region-zones that you would like to use.

Once you have made a decision on that, you could either use the Google Cloud Platform Pricing Calculator to estimate your total project costs, check pricing by the product SKU or see the estimated costs of your VM instance when you create them in the Google Cloud Platform Console.

Please note, that the estimated cost provided by Google Cloud Platform Pricing Calculator is for discussion purposes only and are not binding on either you or Google.

Your actual cost may be higher or lower than the estimate. The following Google Compute Engine Pricing page provides details on how VM instances are billed and how you can save cost on our monthly pricing using the automatic sustained use discounts

Answer 2

There is already an accepted answer but I think it's important to add- this is a complicated problem, with lots of pricing paths and ways to optimize cost. For non-trivial deployments, costs of a naive vs optimized architecture can vary by 1-2 orders of magnitude. Start to wrap your team's head around devoting resources to architecture and cost optimization, it will really make a difference.

Case in point- for running many, many small programs, google offers a product called kubernetes, which can schedule and configure sharing of the resources of individual machines between many individual programs, with fault tolerance and other quality of service attributes. The product is free but requires a distinct kind of architectural expertise.