On Wed, Oct 05, 2005 at 11:15:58PM -0700, william(at)elan.net wrote:

That reminds me. If you remember the whole thing started with that L3 complains that Cogent is trying to steal its customers. I kind of checked and it appears Cogent is after dialup/dsl/cable ISPs who as you can guess have absolutely opposite traffic ratio to typical hosting provider that uses cogent. Obviously this extra traffic does not cost Cogent anything (even if its not peering but transit) and allows it to level its in/out ratio.

Inbound traffic doesn't cost them anything? That old adage only applies to end user transit purchasers who have doing extra outbound and thus have "free inbound" under the "higher of in or out" billing. For folks operating an actual network, the bits use the same resources as traffic in the opposite direction, and thus "cost" the same. The only reason Cogent gives out free or absurdly underpriced inbound transit is ratios.

Now going back to it L3 considers that by offering them connectivity at almost no cost Cogent is dumping - but L3 did the same to get those customers under their contracts some years ago (also in order to even its ratio) and besides that I've heard several times from smaller ISPs (see discussion on isp-bandwidth year or two ago) that they are willing to provide transit for dialup & dsl ISPs at no charge (and I think I know couple cases where that is true) so they would have better ratio for peering. Now Cogent is also offering free transit for single-homed L3 customers to spite L3 after depeering - majority of such single-homed transit customers are in fact these dsl/dialup ISPs Cogent is after which is why they were willing to make this offer ...

I know folks who are willing to give away all manner of things, inbound and outbound, for free or low cost, because they have "excess capacity" that they're already paying for and nothing better to do with it. If you're desperate and you're willing to sacrifice long term marketing for short term cash it can be a cute technique, but to quote Vijay, "it does not scale". Besides, if anyone is depeering Cogent now because of their disruptive pricing in the market, they're a couple years late. Speculate all you like, but I suspect there is more to it than that. -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)

On Thu, Oct 06, 2005 at 01:59:01PM -0400, Patrick W. Gilmore wrote:

You are mistaken.

If I sent 100 Gbps outbound and 20 inbound, I can sell 40-60 Gbps of additional inbound for FAR, FAR less than 40-60 Gbps of additional outbound.

Zero cost? Probably not. Trivial cost? Possibly, depends on network.

Patrick, I keep telling you, you are not an ISP. :) Yes clearly there is SOME reduction in equipment cost at the edge, you need to buy fewer peering and transit ports if there is available capacity on a full duplex circuit in the opposite direction. You may also see some savings on the customer edge where you are utilizing the extra capacity in the opposite direction on trunk ports out of your aggregation layer. Unfortunately in the core traffic is traffic, and you usually don't see such an obvious "but I have this extra capacity in the other direction" pattern. The opex cost of hauling the bits that other folks hot potato onto you is going to quickly negate the capex cost of the equipment. I know you don't deal with this, since as we've already established you are not an ISP, but the cost of longhaul circuits (even very large and well negotiated ones between major cities on major routes) is huge. The cost per meg to get a bit from one side of the US to the other is roughly equal to or above what people are selling transit for per meg these days, and in many cases that doesn't take into account non-perfect utilization and the need for backup capacity on diverse paths. There is nothing trivial about this cost for an actual network, and this completely different from using a rule of 95th percentile billing to squeeze some extra service out of someone else's network for free. Of course you could always make the argument that since circuit costs are usually fixed, you could sell at any price and still make more money than nothing as long as you have extra capacity. This may make you very popular in the industry for a short time, but eventually you will hit a brick wall where you can't afford to buy more capacity on the revenues you are generating. A visit to your local bankruptcy court usually follows quickly.

It doesn't have to scale.

I'm perfectly willing to sell $100K worth of services for $1K worth of cost, knowing I cannot sell $101K because "it does not scale".

Which is why there are a few small networks who don't have extensive circuits and who happen to have some extra inbound capacity available on their transit pipes are selling it for cheap. The concept of "it does not scale" explains why networks are still paying for their bandwidth, even their inbound bandwidth. On the original subject of Cogent, the cost of selling inbound bandwidth is not significantly cheaper than the cost of selling outbound, infact it may actually be more expensive depending on how you crunch the numbers for the fiber and DWDM longhaul capacity.

But I do agree with you on the "couple years late" thing. Putting Cogent out of business will _not_ make prices go up. (And I'm not even sure this will put them out of biz.) In fact, Cogent is not the "lowest cost provider" any more - at least not for bit pushers.

Lots of people out there are emulating Cogent's business model but on a smaller scale in order to deliver a low price/meg number. They're often cutting corners that even Cogent doesn't cut though, and their model only works because a) they're dumping traffic onto peers and transits, and b) they have found transit providers who are as desperate for business at any price as they are. -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)