dc.contributor.author |
Krishnamoorthy, Caucidheesan |
|
dc.date.accessioned |
2024-03-12T10:20:50Z |
|
dc.date.available |
2024-03-12T10:20:50Z |
|
dc.date.issued |
2023 |
|
dc.identifier.citation |
Krishnamoorthy, Caucidheesan (2023) Analysing and Modelling the Latency and Accuracy Trade-offs in Rate Limiting on API-Gateway. BSc. Dissertation, Informatics Institute of Technology |
en_US |
dc.identifier.issn |
20191126 |
|
dc.identifier.uri |
http://dlib.iit.ac.lk/xmlui/handle/123456789/1859 |
|
dc.description.abstract |
"The rate limiting service in API gateways controls request entry by throttling the requests with a boundary. The rate limiting accuracy determines how efficiently rate limiting service works and on whether it allows requests within the throttle count or whether it allows more than the throttle count. Latency on the other hand, is the round-trip time of a particular API call. The accuracy of rate limiting service is defined using spillover error percentage. It is the percentage of requests that rate limiting service allows more than the throttle count.
Requests must be sent to the rate limiting service in order to rate limit requests in API
gateway. The rate limiting service decides whether the requests arrived must be throttled or not. The time taken for the requests to be decided by the rate limiting service adds an additional latency to the round-trip time of the requests. This additional latency can be controlled by introducing a time-out. However, this could result in a degradation in the accuracy of rate limiting. The objective of the project has two folds. First, the author investigates the relationship between accuracy and latency under different workload conditions and timeouts. Secondly, develop a set of machine-learning models that can estimate the accuracy & latency given the relevant parameters.
The analysis of accuracy and latency tradeoff is completed by deriving conclusions based
on the observation of results from performance tests conducted. The conclusions state there is a positive relationship between accuracy and latency and the tradeoff was identified. Voting regressor assembling technique combining the prediction of multivariate linear regressor model and random forest regressor model was used to predict the accuracy and latency trade off. The models have R2 score of 0.85 ± 0.04 with low Mean Absolute Error (< 8.00) and low Mean Squared Error (< 14.00)" |
en_US |
dc.language.iso |
en |
en_US |
dc.publisher |
IIT |
en_US |
dc.subject |
Spillover error percentage |
en_US |
dc.subject |
Latency |
en_US |
dc.subject |
Rate limiting service |
en_US |
dc.title |
Analysing and Modelling the Latency and Accuracy Trade-offs in Rate Limiting on API-Gateway |
en_US |
dc.type |
Thesis |
en_US |