3 ms·
Dropping packets makes load problems worse, because failing users start to send even more traffic.
by jfoutz 12y ago
Dropping packets makes load problems worse, because failing users start to send even more traffic.
- MAGZine 12y agoThe opposite, dropped packets are bad because it slows the Internet down. When you need to retransmit a packet because one is dropped, the network assumes congestion and radically throttles speed to prevent the network blowing up.
- Nacraile 12y agoThis is not true. TCP will retransmit, yes, but TCP congestion control will also reduce the transmit rate in response to lost packets. Overall, proper TCP implementations will back off to approximately their fair share of a contended bottleneck. The application layer sees reduced throughput, not message loss. New TCP connections start with very conservative transmit rates, so an application which is overly aggressive in abandoning slow connections and retrying will not add substantially to congestion. UDP doesn't necessarily have these properties, but anyone who implements a high-bandwidth UDP protocol which isn't TCP congestion control friendly deserves public shaming.
- SoftwareMaven 12y agoExcept between buffer bloat[1] and users spamming the reload button, TCP's congestion control is not going to solve the problem. 1. http://en.wikipedia.org/wiki/Bufferbloat http://en.wikipedia.org/wiki/Bufferbloat
- deeviant 12y agoThat is true. TCP congestion control throttles back the speed in the presence of dropped packets. Thus, dropped packets adversely affect throughput at a magnitude greater than the simple overhead of packet retransmission.