Benefits of RED/WRED

So how do you use RED/WRED at your company? I've done a lot of QOS implementations, and I've always questioned the value of RED with how most companies implement it.

I've heard arguments that it helps prevent TCP global synchronization, but is synchronization actually happening in your network without a RED profile? Most of the research I've read showing global synchronization conducted their testing in a controlled environment with flows having similar characteristics (algorithm, latency, processing time, RTT). I have not seen global synchronization in our production network, nor have I seen it in a lab where we weren't explicitly trying to create it. Reading through some research where different RTTs and different processing times are used, you don't see global synchronization occurring.

What's consistent with a lot of the research I've read through is fairness and latency. RED is fairer to bursty traffic or traffic with long RTT. But for the most likely scenarios, tail-drop works fine or may even be better for smooth non-bursty traffic. In any case, the average of drop probability and throughput between tail-drop and RED are pretty identical. While latency can be far lower with RED, that's because you're essentially lowering the queue depth. This could be ideal in specific scenarios, especially with WRED. But that kind of granularity requires engineers to have an excellent understanding of their traffic flows and needs.

What I often find with most deployments, RED/WRED is implemented without really knowing if it provides any benefit. There are a few scenarios where I could see the benefit, but I don't even know if the benefit is worth the added complexity.

So if you have RED/WRED implemented in your network, how is it implemented and what benefit does it provide? How did you determine the min and max thresholds and drop probability? Have you done testing to show that it actually provides an advantage over just doing tail-drop, or is it all theoretical?