Comments on: Twenty Questions for the FCC http://bennett.com/blog/2008/07/twenty-questions-for-the-fcc/ A personal blog Sun, 11 Sep 2011 14:57:16 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Richard Bennett http://bennett.com/blog/2008/07/twenty-questions-for-the-fcc/#comment-5575 Richard Bennett Sun, 03 Aug 2008 21:48:35 +0000 http://bennett.com/blog/?p=4691#comment-5575 Comcast has a point-to-multipoint network, where the most severe bandwidth constraint is on the upstream side. Consequently, they need to deploy a system that enforces fair queuing among bandwidth requestors. That's tricky in DOCSIS, as a scheduling decision needs to be made in real time, but it's not impossible. Comcast has a point-to-multipoint network, where the most severe bandwidth constraint is on the upstream side. Consequently, they need to deploy a system that enforces fair queuing among bandwidth requestors. That’s tricky in DOCSIS, as a scheduling decision needs to be made in real time, but it’s not impossible.

]]> By: Brett Glass http://bennett.com/blog/2008/07/twenty-questions-for-the-fcc/#comment-5574 Brett Glass Sat, 02 Aug 2008 20:24:59 +0000 http://bennett.com/blog/?p=4691#comment-5574 The problem is more complex than that. There are actually TWO constraints on downstream equipment, whether it's wired or wireless: bit rate and packet rate. (Due to the overhead in processing a packet, the packet rate becomes important when there are many small packets sent very quickly -- such as by a BitTorrent tracker or seeder or by GNUtella's UDP discovery mechanism. The latter can flood a network with requests for weeks or even months after the machine that was running GNUtella was shut down.) It's also worth remembering that if a user consumes constant bandwidth, it's a constant and large cost to the ISP. At $100 per Mbps per month, an increment of 128 Kbps is $12.80. Our ISP makes less than $5 per customer per month, so that's enough to cause a net loss on the customer. The problem is more complex than that. There are actually TWO constraints on downstream equipment, whether it’s wired or wireless: bit rate and packet rate. (Due to the overhead in processing a packet, the packet rate becomes important when there are many small packets sent very quickly — such as by a BitTorrent tracker or seeder or by GNUtella’s UDP discovery mechanism. The latter can flood a network with requests for weeks or even months after the machine that was running GNUtella was shut down.)

It’s also worth remembering that if a user consumes constant bandwidth, it’s a constant and large cost to the ISP. At $100 per Mbps per month, an increment of 128 Kbps is $12.80. Our ISP makes less than $5 per customer per month, so that’s enough to cause a net loss on the customer.

]]> By: Richard Bennett http://bennett.com/blog/2008/07/twenty-questions-for-the-fcc/#comment-5573 Richard Bennett Fri, 01 Aug 2008 22:14:20 +0000 http://bennett.com/blog/?p=4691#comment-5573 Depends how many people are doing it. If you have 20 on one cable segment, you're going to get a lot of DOCSIS collisions, and if you are one of a small number of providers of a rare file piece, you're going to have to field a large number of socket opens, each requiring a response. Depends how many people are doing it. If you have 20 on one cable segment, you’re going to get a lot of DOCSIS collisions, and if you are one of a small number of providers of a rare file piece, you’re going to have to field a large number of socket opens, each requiring a response.

]]> By: Wes Felter http://bennett.com/blog/2008/07/twenty-questions-for-the-fcc/#comment-5572 Wes Felter Fri, 01 Aug 2008 21:46:45 +0000 http://bennett.com/blog/?p=4691#comment-5572 Hours-long seeding at 128kbps can't cause that much congestion on a 10-20Mbps link, although I guess the transit costs could be substantial: 13GB/day. Hmmmm. Hours-long seeding at 128kbps can’t cause that much congestion on a 10-20Mbps link, although I guess the transit costs could be substantial: 13GB/day. Hmmmm.

]]> By: Richard Bennett http://bennett.com/blog/2008/07/twenty-questions-for-the-fcc/#comment-5571 Richard Bennett Thu, 31 Jul 2008 03:02:35 +0000 http://bennett.com/blog/?p=4691#comment-5571 Brett has a lot of custom-built and open source tweaks in his wireless routers, and I'm sure rate control is among them. I think part of the issue is that wants to allow occasional FTP and HTTP uploads but not persistent, hours-long random seeding sessions. Brett has a lot of custom-built and open source tweaks in his wireless routers, and I’m sure rate control is among them. I think part of the issue is that wants to allow occasional FTP and HTTP uploads but not persistent, hours-long random seeding sessions.

]]> By: Wes Felter http://bennett.com/blog/2008/07/twenty-questions-for-the-fcc/#comment-5570 Wes Felter Thu, 31 Jul 2008 01:14:35 +0000 http://bennett.com/blog/?p=4691#comment-5570 I wonder if any WISPs have considered something like a 128kbps,10MB token bucket; it's neutral, it should satisfy light users, and it limits hogging. I wonder if any WISPs have considered something like a 128kbps,10MB token bucket; it’s neutral, it should satisfy light users, and it limits hogging.

]]>