C1W3: tech components used to implement complex architecture

In the video on Meta-data, data provenance and lineage , a complex architecture with multiple models is shown, compare screenshot below:

I know this is just an intro course but I am curious as to how such a system would be implemented and what technologies could be involved. Such as:

  • are the different models exposed as web APIs or accessed otherwise?
  • is there a central orchestration layer or does every endpoint send the output to the next model? Where to encode the logic for the routing?
  • would this be deployed to a K8 cluster, a spark cluster or what other type of infrastructure could be used?

I assume the answer is often it depends, but then it would be great to know what it depends on. Thanks

Hi @lorenzwalthert and welcome to the course!

As you noticed, there are many ways to build such a system. In the following course of the specialization, you will have a chance to work with Google ai-platform, which you can use to deploy your model (and expose it with REST API (recommended way by Google documentation )

And it depends on many factors like what latency and throughput do you expect from the system, it will affect all the architects. If you serve your model inside a company with a few hundred employees then it may just work with one server on-premise. And to find which is good, I think you need to do a lot of performance tests to find out which architect fits your case.

In summary, the followings are a few factors that I think you need to consider when architect your system:

  • How many users you need to serve?
  • How much budget you have?
  • How many members in your team manage the system?
  • How is the performance requirements: latency, throughput?
  • Is it public or private inside the company (you need to work more on security if it is public)?

Hope it helps,
Cuong

1 Like