Rate limiting algorithms

In this section, we will explore the different algorithms available for rate limiting using the Borrow API. If you’re more of a visual learner we recommend visiting this article. 

The previously mentioned article only covers 3 of our 4 algorithms, that’s because we’ve developed the borrow algorithm, even though it’s less of an algorithm, more of just synchronous rate limiting.

Fixed Window

The fixed limiter is simply an algorithm that limits the number of requests to a fixed number on the clock.

For example, if you set the limit to 10 requests per minute, the first 10 requests will be allowed, and the next requests will be blocked until the next clock minute. In the fixed window algorithm, it doesn’t matter if the 10 requests were made 1 second before the next minute or however long before the next minute. In either scenario, it will reset in the next minute on the clock.

You can use the fixed window algorithm by specifying the limiter type to "fixed".

Sliding Window

The sliding window algorithm is a more sophisticated approach to rate limiting that allows for a more fair distribution of requests over time. In this algorithm, the limit is applied over the time window at the time of the request rather than a fixed clock interval.

For example, if you set the limit to 10 requests per minute, the first 10 requests made within any 60-second window will be allowed.
Let’s say the user makes a request at 00:00:30, that means their 10-request limit will reset at 00:01:30, rather than at 00:01:00, as in the fixed window algorithm.

You can use the sliding window algorithm by specifying the limiter type to "sliding".

Token Bucket

The token bucket algorithm is the most flexible of our rate limiting algorithms.
In this algorithm, tokens are added to a bucket at a rate you specify (interval and tokensPerReplenish), and each request consumes a specified amount of tokens from the bucket (tokensCost).

After the maximum amount of tokens is reached (maxTokens), the requests will be blocked until the user has enough tokens. Likewise, if the user only has 70 tokens, and the request consumes 80 tokens, the request will be blocked until enough tokens are replenished.

Borrow

The borrow algorithm is a synchronous rate limiting algorithm that allows you to limit requests without the need for a time window.
In this algorithm, you call the API when a user has made a request, and then you call it again when the user has finished the request.

In the time between the two calls the user is blocked from making any requests. You’ll also need to specify a timeout parameter to ensure that if you forget or are unable to call the API again the user will be automatically unblocked after the timeout has expired.