From the article:
"
Network equipment has used this same technique for 40 years, essentially putting arriving packets into an output queue and if it fills up, dropping packets. The packets are dropped randomly, which is the source of most of the unfairness attributed to TCP. Random drops cause some flows to stall while others flourish.
What is really necessary is to detect just the flows that need to slow down, and selectively discard just one packet at the right time, but not more, per TCP cycle. Discarding too many will cause a flow to stall â" we see this when Web access takes forever.
"
sounds like we need to engineer networks to not drop packets, and when congested to the point where packets need to drop (there is no room in the software queue), we need to selectivly drop the packets. we do this, and packets are going to drop regardless. Working with flow-information isn't anything special. With this we can randomly drop flows to prevent tcp from creating congestion when they all synchronize up at the same time. Only traffic that doesn't meet certain criteria can have this method applied, other traffic can be given priority in the queue so it never drops. Also we can analyze layer seven data to classify the packet in the queue. In addition to these methods the flow-based information wouldn't provide much benefit at all, if any. There is simply not a need for it, and that's why the interenet isn't using this technology. Maybe a couple people use his product, good for them I guess, but I'm glad the internet isn't running on this company.
If we put a comparision of the current technology in place, it's design, and what his products could do, you would see we are in the right hands. This company is competing with CEF, NBAR, WRED.. the network is already engineered, I didn't see that mentioned in the article. Although I did see four key problems, which weren't addressed in comparion to what we are doing now, and why we are doing it.
I would compare a system of his equipment powered by his network to Cisco's
Slashdot Mirror
From the article: " Network equipment has used this same technique for 40 years, essentially putting arriving packets into an output queue and if it fills up, dropping packets. The packets are dropped randomly, which is the source of most of the unfairness attributed to TCP. Random drops cause some flows to stall while others flourish. What is really necessary is to detect just the flows that need to slow down, and selectively discard just one packet at the right time, but not more, per TCP cycle. Discarding too many will cause a flow to stall â" we see this when Web access takes forever. " sounds like we need to engineer networks to not drop packets, and when congested to the point where packets need to drop (there is no room in the software queue), we need to selectivly drop the packets. we do this, and packets are going to drop regardless. Working with flow-information isn't anything special. With this we can randomly drop flows to prevent tcp from creating congestion when they all synchronize up at the same time. Only traffic that doesn't meet certain criteria can have this method applied, other traffic can be given priority in the queue so it never drops. Also we can analyze layer seven data to classify the packet in the queue. In addition to these methods the flow-based information wouldn't provide much benefit at all, if any. There is simply not a need for it, and that's why the interenet isn't using this technology. Maybe a couple people use his product, good for them I guess, but I'm glad the internet isn't running on this company. If we put a comparision of the current technology in place, it's design, and what his products could do, you would see we are in the right hands. This company is competing with CEF, NBAR, WRED.. the network is already engineered, I didn't see that mentioned in the article. Although I did see four key problems, which weren't addressed in comparion to what we are doing now, and why we are doing it. I would compare a system of his equipment powered by his network to Cisco's