Ordinarily, I'd write this up as news, but it be a tad over my head, still interesting none the less.
http://arstechnica.com/tech-policy/...ferbloat-and-the-network-buffer-arms-race.ars
In a nutshell, the large buffers on home routers and switches can cause high latencies for other traffic when they become full. Normally you'd think that large buffers are a good thing, smooths out traffic, etc, but when the buffer can hold up to 2 or 3 seconds (or even 10) of traffic during a download, and you want to browse the net... then you have to wait for the buffer to clear before your web packet can get through...
Now, this is normally when Quality of Service (QoS) is brought up, and it indeed does mitigate the problem. However, it wouldn't really be needed if the buffers were smaller (not too small mind), since the packet would be dropped and resent instead of held up for several seconds in a buffer. Problems with TCP transfer protocols with Flow rate and such.
I guess there is more to the humble balancing act of the home router than meets the eye. Rather strange to think about this in a different way... a slow connection to a website with a download running in the background may not be the result of bandwidth saturation, but of a full buffer... who'd of thought... I guess that explains why long distance downloads can result in the same delays, even though you are not maxing out your connection.
http://arstechnica.com/tech-policy/...ferbloat-and-the-network-buffer-arms-race.ars
In a nutshell, the large buffers on home routers and switches can cause high latencies for other traffic when they become full. Normally you'd think that large buffers are a good thing, smooths out traffic, etc, but when the buffer can hold up to 2 or 3 seconds (or even 10) of traffic during a download, and you want to browse the net... then you have to wait for the buffer to clear before your web packet can get through...
Now, this is normally when Quality of Service (QoS) is brought up, and it indeed does mitigate the problem. However, it wouldn't really be needed if the buffers were smaller (not too small mind), since the packet would be dropped and resent instead of held up for several seconds in a buffer. Problems with TCP transfer protocols with Flow rate and such.
I guess there is more to the humble balancing act of the home router than meets the eye. Rather strange to think about this in a different way... a slow connection to a website with a download running in the background may not be the result of bandwidth saturation, but of a full buffer... who'd of thought... I guess that explains why long distance downloads can result in the same delays, even though you are not maxing out your connection.
Last edited: