กลยุทธ์ตัวนับหน้าต่างเลื่อน

Welcome to our lesson on the Sliding Window Counter (SWC) algorithm!

This algorithm is a clever way to implement rate limiting. It aims to offer better accuracy than the simple Fixed Window Counter, while being more memory-efficient than the precise Sliding Window Log.

Recall that the Fixed Window Counter can suffer from a 'burst' problem. If a user makes requests right at the end of one window and then again right at the start of the next, they can effectively double their allowed requests in a short period.

The Sliding Window Counter helps to mitigate this issue.

Instead of just looking at the current fixed window, SWC looks at two fixed windows:

• The current window.
• The previous window.

It then combines their counts using a weighted average to estimate the true request rate over a 'sliding' period.

The 'sliding' effect comes from how we weight the previous window's count. We calculate an overlap percentage based on how far we are into the current window.

For example, if our window size is 60 seconds and we are 30 seconds into the current window, 50% of the previous window is still 'relevant' to our current sliding view.

To apply the Sliding Window Counter, you need to track a few pieces of information:

• The total request count for the previous fixed window.
• The total request count for the current fixed window.
• The current timestamp (to determine how far into the current window we are).
• The defined window size (e.g., 60 seconds, 1 minute).

Let's use an example:

• Rate Limit: 10 requests per minute.
• Current Time: 30 seconds into the current minute.
• Previous Minute's Count: 8 requests.
• Current Minute's Count: 3 requests so far.

How many requests have we 'used' in our sliding window?

Here's how we calculate the estimated count:

1. Overlap Percentage: We are 30 seconds into a 60-second window, so 30/60 = 0.5 (or 50%).
2. Weighted Previous Count: The previous window's count (8) is weighted by (1 - overlap percentage). So, 8 * (1 - 0.5) = 8 * 0.5 = 4.
3. Estimated Total: Add the weighted previous count to the current count: 4 (weighted prev) + 3 (current) = 7.

So, 7 requests are estimated, leaving 3 requests remaining.

This simple Java code snippet demonstrates how to calculate the estimated request count based on the current state of two windows.

Try running it to see the calculation in action!

The Sliding Window Counter offers several advantages:

• Improved Accuracy: It provides a better approximation of the true rate than Fixed Window, especially around window boundaries.
• Memory Efficiency: Unlike Sliding Window Log, it doesn't need to store every request timestamp, making it less demanding on memory.
• Better Burst Handling: It reduces the chance of allowing excessive bursts compared to the Fixed Window algorithm.

While powerful, the Sliding Window Counter is still an approximation. It's not perfectly accurate like the Sliding Window Log.

It can still allow slight overages at window boundaries, though significantly less than a purely Fixed Window approach. For high-precision requirements, the Sliding Window Log might still be preferred, if memory allows.

Let's test your understanding of the Sliding Window Counter calculation!

Great job! You've learned about the Sliding Window Counter algorithm.

• It combines counts from two fixed windows to approximate a sliding window.
• It's more accurate than Fixed Window and more memory-efficient than Sliding Window Log.
• It calculates an estimated count using an overlap percentage.

Next, we'll compare all the algorithms you've learned to understand their trade-offs.