Amazon SageMaker vs. AWS Lambda

Amazon Sagemaker suits well in areas of data science and Machine learnings where medium to high-volume data is to be used for analysis. For a lean and platform agnostic deployment, it provides kubernetes integration to containerize the solution and deploy on any platform. It is one of the best solution for technical users for training Machine Learning models.

Incentivized

Scenarios where AWS Lambda is well suited: 1. When we need to run a periodic task few times in a day or every hour, we may deploy it on AWS Lambda so it would not increase load on our server which is handling client requests and at the same time we don't have to pay for AWS Lambda when it is not running. So, overall we only pay for few function invocations. 2. When some compute intensive processing is to be done but the number of requests per unit of time fluctuates. For example, we had deployed an AWS Lambda for processing images into different sizes and storing them on AWS S3 once user uploads them. Now, this is something that may happen few times every hour on a particular day or may not happen even once on other days. To handle this kind of tasks AWS Lambda is a better choice as we don't have to pay for the idle time of the server and also we don't have to worry about scaling when the load is high. Scenarios where AWS Lambda is not appropriate to use: 1. When we expect a large request volume continuously on the server. 2. When we don't want latency even in case of concurrent requests.

Incentivized

• SageMaker is useful as a managed Jupyter notebook server. Using the notebook instances' IAM roles to grant access to private S3 buckets and other AWS resources is great. Using SageMaker's lifecycle scripts and AWS Secrets Manager to inject connection strings and other secrets is great.
• SageMaker is good at serving models. The interface it provides is often clunky, but a managed, auto-scaling model server is powerful.
• SageMaker is opinionated about versioning machine learning models and useful if you agree with its opinions.

Incentivized

• AWS Lambda is a welcoming platform, supporting several languages, including Java, Go, PowerShell, Node.js, C#, Python, and Ruby. And if you need to deploy a Lambda function in another language, AWS offers a Runtime API for integration.
• We really appreciate how AWS Lambda is always-on for our functions, with only a brief "cold-start" waiting period the first time a function is called after being dormant.
• In addition to only generating costs when it's actually being used, AWS Lambda really puts the "serverless" in serverless architecture, offering turnkey scaleability and high availability for our code with zero effort on our part.

Incentivized

• Searching and descriptions can be easier to read and interpret.
• Training modules and customer service training representative could make on boarding employees easier.

Incentivized

• The UI and Developer experience is not so great. IF you use an abstraction like Serverless Application Model (SAM), things get pretty easy, but it's still AWS UI/DX you're working with after that (which is to say, not their strength).
• Documentation is always a mixed bag. Sometimes it's just easier to google your specific problem and see how others have solved it. This can be much faster than trying to find an example that may or may not be there in the documentation (which oftentimes has multiple versions and revisions).

Incentivized

It is very easy to get started with AWS Lambda and create your first function. The user interface makes it easy to add AWS services to be inputs or outputs to the function, meaning it can be configured in many different ways for different needs. This makes it ideal for various scenarios in AWS.

Incentivized

As this is a product where a great part of errors can be at the source code level, AWS support team doesn't dive that further. I mean they don't evaluate problems more complex related to your code, [which] is totally understandable, but this make[s] debug process more tough and painful.

Incentivized

We have not invested in another machine learning software at this time and so far this has proved very successful with our machine learning teams. As mentioned, I am training these individuals simply on the fundamentals of the software and using it/customizing it for their needs. It has been very easy to do this and has gotten great reviews across the organization so far.

Incentivized

It's fine, it works as the others would have, except EC2. We are migrating back to EC2 for dedicated compute because we have scaled to a point where we have consistent traffic. The tradeoff of maintaining infrastructure in-house outweighs the benefits of moving quickly through our roadmap.

Incentivized

• Using SageMaker, we can truly implement 'fail early, learn fast,' using an on-demand server for training.
• It also saves your money from investing in a physical server for very rare use.
• However, the pricing is high, but it will cost you only for what you use.

Incentivized

• We have simplified log fiie ingestion using Lambda functions. The return has been less time worrying about getting logs from source to ingestion; one the process is in place the team is nearly 100% hands off.
• We have begun taking a more API focused approach by using API Gateway as the interface to business processes and Lambda as the back end compute. Moving away from server based back ends places us on a path to reducing overall spend in compute costs.
• Lambda functions allow us to easily interface with third party services through APIs. This simplifies access management since the function can be granted permissions and access to the function can be gated with API keys and other authentication methods.

Incentivized