[Costless] - Revised one with transcript - p 21~27 page

 21 page

 

We can define models in many aspects. For the resource model, we think the AWS Lambda resource model as f with only one variable as a requesting memory. Other options are the default. 

 

 

22 page 

For Data model - The data sent to the Lambda function is in the form of a request encoded in JavaScript Object Notation (JSON) format. The above request is retrieving images from S3.

23 page

 

Model the workflow in AWS Lambda as a directed-acyclic-graph (DAG) Gf = (Vf, Ef) of functions

Vertices represent each function 

Edge represent workflow dependency like f1’s output is input of f2.

24 page

When profiling each function, we consider 4 things

 

1. Execution cost - it could be executed on cloud or edge

2. Transmission time from edge to cloud.

3. maximum memory consumed by the function

4. Scheduling delay - it’s time between receiving the request and starting to execute the function.

25 page

To sum up, the price model like above as follows:

 

It considers the placement of function, cloud or edge

 

fixed cost for edge device

 

price of executing the function on the cloud with designated memory and executing time

 

The price per 1GB memory and 1 sec of execution time

 

the executed transition between functions.

26 page

 

For Execution time model - It’s adding executing time on edge and cloud, transmission time. 

 

 

27 page

 

To sum up, It’s about minimizing the price where execution is time is less than threshold execution time