Just as I thought...give 'em as much as they use!

I mentioned yesterday that I was suspicious of Robb Topolski's definition of acceptable network management practices.  Well, just as I suspected, Robb justified my concern right here in a comment to my post yesterday.

Yes, I approve of Network Management – the old fashioned kind – the kind that says you stay ahead of demand by upgrading your networks on time and you don’t oversell your bandwidth beyond all reasonableness. 

That's a great idea but it naively doesn't address the realities of running a real network on today's Internet.  Instead, it sounds like this is an intellectual debate.  To the best I can tell, people who think we should build whatever a few customers can consume, have never taken a call from a customer, received a daily usage report, designed a node split and split it, or sent out a bill. 

In other words, they have never managed a network.

I'd love to just build as much as anyone can consume.  But when it comes to upstream usage, I have some experience and it simply won't work.

How do I know that?  Because two years ago, we effectively managed our network by limiting everyone's upstream speed at a fairly low level.  We decided to offer a nearly tripling of upload speeds to all customers.  We cranked it up from 386 to 1 meg, becoming one of the fastest ISP's in the nation. 

Within three days, we found that our drain points to the Internet were clogged with uploads.  P2P software all over the world "learned" that there was a new very fast uploading ISP out there and more and more requests came our way. By day 3, fewer than 5% of our customers were utilizing 80% of our capacity, feeding P2P packets to anyone, anywhere in the world. 

Sadly, the only thing that stopped more requests from coming in was the fact that we were too clogged to handle any more.  And when that happened, everyone's Internet experience slowed to a crawl, no matter what they're doing.  Luckily, we were prepared to begin new network management practices and we accelerated doing so.

But, if we did what Robb suggests above -- we simply build more -- it would only reopen our network to more incoming requests from all over the world to feed more uploads which, within a day or two would simply put us back to the same clogged condition. 

Instead, we build enough capacity to give the P2P community a hugely disproportionate amount of upstream capacity.  It's simply not unlimited.  And, by the way, our customers have a pretty good P2P experience with our current management practices. 

How do I know?  They tell me.

So go ahead and poke around.  Theorize all you want. 

But I have come to two very important conclusions about this subject.

1) Our customers like their current broadband experience, including P2P enthusiasts and

2) Some of the "fairer" alternatives to our current management practices will actually be more intrusive to more customers, including P2P users.

If the FCC decides to find that our current practices violate consumers' freedom to go anywhere on the Internet, it will be a work of fiction.  We're not blocking anything, we're just helping people gain access to it all fairly.

Comments

[ARGO]

I hate to say it but... I believe it really falls back to lack of competition with cable.

If free market competition would play out then supply and demand would push technology and all this debate about upstream bandwidth limiting would be moot. "that's my outside take"

I do enjoy using 20.0 Insightbb btw :)

[funchords]

Michael,

Opening up the cable modems is not upgrading the network. For example:

The village well is running dry. To solve the problem, do you:

a. dig additional deeper wells
b. give the villagers bigger buckets?

You apparently picked "b" and got an undesirable result.

Picking "a" would involve node splitting, and may be expensive. In the end, you would either have fewer subscribers sharing the same-sized bandwidth pool or the same number of subscribers sharing a larger sized pool.

Choosing "b" is inexpensive, doesn't change the number of subscribers or the pool, but has the side effect of making user applications believe that the upload pipe is now bigger when, on the average, it has remained the same size.

I'm enjoying this exchange, I hope it can continue in good tone.

I do think we're on the verge of teaching something terrific here to the users, and that is learning that Cable Internet, as currently deployed, is inferior to Telco for P2P uploads. Cable is superior for bringing in the downloads, but was not built for uploading in mind.

If we can admit that, we can do a few things:
1. Change users' expectations. Instead of configuring their P2P applications to use 80% of their uplink speeds, they should use much less (say 20%).
2. Allow consumers to choose the service that is best suited for them. Heavy downloaders from web, ftp, and "public" P2P trackers should stick with cable. Heavy uploaders, including users of "private" P2P trackers should move to another service or augment their service (I, personally, have both Cable Internet and DSL here at the house).

Heavy uploads is not what Cable is currently about. Hopefully the industry can change that as it moves more to SDV and DOCSIS 3 in the coming years. (Cox Cable is even hinting about FTTH!)

Looking back 10 years, perhaps this was the thought that preceded ISPs providing hosted personal web pages. If people hosted their own web pages instead, it could overwhelm the asymmetric model Cable relied upon when designing its systems.

Fast-forward to today, ISPs can host technology known as "seedboxes" or "remote servers" (two different things but both solutions worth investigating) to help take the pressure off of DOCSIS upstream. By doing this, the upstream box handles the P2P "give-back" on the IP side and the data coming to the user only crosses DOCSIS in the downstream direction.

As for Deep Packet Inspection, I'm not anti-DPI, but it doesn't belong interfering with users' desired communications on the net. That's not your role as the ISP. It's not your data, and you have no right to change a byte of it. But rather than manage the inner workings of the very applications that users use, persuade them by presenting a more attractive alternative.

I appreciate your blog and the tone of it. Perhaps we can keep this conversation in the "light" instead of in the "heat." Two rather incendiary elements are Richard Bennett and George Ou, both of which have elected to promote bias over technical facts, and both have elected to attack me rather than debate me.

Robb Topolski

[Chris Buechler]

Everyone with any network sense understands you can't let customers run wild on residential connections and provide good service at reasonable prices. The means of resolving the problem of excessive use vary. Personally, I'd be happier if Insight didn't "manage" my traffic whatsoever, but rather offered *reasonable* usage-based billing. Reasonable is key, given what some other ISPs are now implementing with usage caps providing a low amount of transfer and charging exorbitant amounts for excess usage.

[ARGO]

If any network cant handle it's users traffic now how can it handle DOCSIS 3.0 level traffic?