Bufferbloat, The Unknown Phenomenon

Have you thought about the speed of downloading and uploading and their impact on each other? The two seem to be separated from each other.

Jim Gettys is an American computer programmer at Alcatel-Lucent in Bell Labs. Until 2009, Gettys was vice president of software for the One Laptop for Every Child project.

At home, he was uploading a large file to his work server. Meanwhile, her children came to her office and complained about the low speed of the Internet. Surprisingly, his uploading work could affect his children’s download speeds, so he began to investigate, and his research led to the discovery of a malicious and unknown phenomenon called bufferbloat.

This problem is directly related to how the TCP protocol works and how the buffers are managed in the network, which is not yet fully understood how it works.

There is a widespread belief that the loss of packages on the Internet is always a negative side effect. But the fact is that packet dropdowns are essential for TCP to work properly.

Many believe that this method is to correct almost any performance error and to increase bandwidth.

To reduce the number of missing data packets on the Internet, developers have increased the size of network buffers several times in a row. Although this increases response time, it has little effect on overall performance. Consequently, small essential packets, such as those used in DNS, may be trapped in the buffers of larger packets related to file transfers or other transfers.

There is a conceptual problem associated with buffer management, tests, and even guides that often identify buffers as a small part of memory. Buffers can often hold hundreds or even thousands of packages at a time. These are not limited to network devices. They can also be seen in the network card drivers and any ports of end devices.

We need to consider two issues:

1. How does the sender find out if there is congestion in the network or not?
2. When it finds out that there is congestion, how does it slow down?

what are the Reasons for congestion?
The sender speed is high and the middle router buffer is full and there is a delay or The buffer fills up and loss occurs and it is forced to constantly re-transmit traffic.

In the TCP model, the network does not tell the sender at all that There is congestion in the link and it just does only through communication between the sender and The recipient notices that there is congestion.

They are responsible for monitoring the proper performance of TCP density control operations. By combining packets in queues, small essential packets such as DNS responses are not trapped in long queues. In other words, using this method, how to deal with large and small packages will be done more equally. Numerous studies have shown that the benefits of using fq-codel are far greater. fq-codel is used only in the latest Linux distributions.

Scroll to Top