An ISP for the 21st century.

Tue Jul 22 03:50:00 -0700 2008
manage

Back in about '94 there were lots of people about into BBS and Fido, and there were lots of new computer owning people wanting to get online and connected, there was this new WWW thing see...

So I met up with this local guy called Mark, he had borrowed about half the cost of a small house and bought a couple of racks of modems, a server or two and pipe into the backbone, such as it was. At the time I had some spare capital and I could have thrown my lot in with him, but his service was £20 a month to dial into his new, fast, 14.4k modems, plus the phone bills at 3.5 pence per minute, plus when you finally got on to the "WWW" there wasn't any content there to speak of, so I looked at all this and decided that while I might be a customer of his, I wasn't going to buy into it as it was obviously doomed.

I was wrong and he was right, mainly because I failed to see that his main targeted customer was business, the other local ISP who targeted punters lasted about two years.

So basically in those days an ISP was someone you paid money to in order to log-in to their systems and thus connect to the wider internet, oh, and you got 10 megabytes of web space too, and an email account.

So the ISP was really just a glorified BBS node, and that was all she wrote, and your phone bills were still IRO 500 quid per quarter....

Fourteen years have passed, and today the ISP is quite different.

Back in those days the eventual head of what became one of the country's largest ISP's said to me "the internet is all about copyright violation anyway" and he was OK with that, but start screwing with the network as a whole by dumping malware or suchlike and your account was history and so were you... sure you got a warning, but it was a serious one, an "or else" and once sanctioned you were basically cut off.

Since those days the role of the ISP has changed, and indeed the legal status too (of which changes I have played my part in) but the biggest change is the basic business model.

In those days you bought or leased a pipe, and then you turned around and tried to sell it at a profit, and the sweet spot commercially was around ten to twelve customers per lines in and modems, (back then phone lines automatically disconnected you after 2 hours) any more than that and people got frustrated unable to log-in, any less and you had under-utilised capacity, but the basic deal was buy X, cut it into slices and sell the slices for X plus profit margins.

Today the business model is first you get your customers, then you put those slices together, then you take out your profits and with whatever is left you buy a pipe, and since connections are now "always on" we now call this a "contention ratio" on the customer side and "oversubscription" on the ISP capacity side. (Technically contention ratio is how many customers share a specific pipe to the ISP, in the old days, "modem sharing")

So contention ratios started out, naturally enough, at about ten or twelve to one, but times passed, and today contention ratios of between 20:1 and 50:1 are not uncommon, plus the sum of all customers also wildly exceeds upstream pipes from the ISP.

The other thing that changed was routing redundancy, back then this early ISP had a main T1 connection, cost a lot of money, but it also had dual bonded 128k ISDN as a backup, and a single 64k ISDN as a third backup, with priority to port 25/110 for email, and when the main pipe was up these "backchannels" were used for the less important stuff like syncing the ISP's internal NNTP pool and suchlike.

With the change in business model all this stuff went, when you take your profits out and use what's left to buy a pipe then all these things get the chop, local NNTP, cut it, local FTP, cut it, redundant routing, cut it, round robin DNS, cut it, redundant hardware, cut it, perhaps most significant of all, network abuse attack dogs, cut it, but after a while even this isn't enough.

So all that is left is throttling, euphemistically called network neutrality or traffic shaping or QoS if you happen to be the one holding the tap, but, it is throttling plain and simple.

So the brand that is Virgin media sells me a 20mbit down 768k up pipe for 40 quid a month, well, no, that isn't true, they took over the 10m/512k blueyonder pipe, and "upgraded" it by opening the tap a bit, thus overselling the resource, then proceeded to continue to oversell, so that irrespective of the time of night or day I am lucky to see a sustained 450k down out of my fat 20 mbit pipe.

Which isn't that surprising, because a little digging reveals that there are, including me, 68 cable modems sharing the same node as me, so the actual contention ratio here has gone from 7:1 with Blueyonder to 68:1 with Virgin.

And so, just before the NTL Business broadband (20mbit down, 1 mbit up, guaranteed no traffic shaping, guaranteed better than 12:1 CR, guaranteed SLA, guaranteed phone, all for the same money as Virgin) guys turn up to do the site survey the postman arrives...

Junk mail, including one huge glossy "ooh, shiney" red brochure from Virgin, the guys who have oversold my 20 mbit connection down to 25% of nameplate already, and what does it say?

My 20 mbit connection is cheaper than I am paying now. £10 p.c.m. for the first 6 months, thereafter £20 p.c.m. for a 20 mbit (ooh, "up to") connection, how could anyone possibly not want a piece of this bargain of the century.

So if I thought the traffic shaping (what traffic shaping? have you tried rebooting your computer sir?) was bad NOW, wait a few months, except I will be long gone as a customer... same fibre though...

Which brings us to the point and title of the article, an ISP for the 21st century, what do we want?

Well, I dunno what you want, but I know what I want from a 21st century ISP.

  1. I want a nameplate speed up and down that bears some resemblance to actual throughput.
  2. I want a fixed IP as standard.
  3. I want no restrictions on what I connect at my end, I am buying a pipe, nothing more or less.
  4. I want no traffic shaping, network neutrality or QoS imposed on me.
  5. I want a smart box***, not a cable or adsl modem.

By a Smart Box*** I mean a box that performs the modem function, a box that povides DNS for my subnet, a box that provides a firewall, and a box that provides traffic shaping for my subnet, all in one, with a web based admin interface.

By all means, ship the box with a default set of parameters in order that port 25 / 100 is top priority and firewalled to 5 recipients per email and 100 emails an hour, by all means, ship the box with a default set of parameters that limits torrent ports to 10% of my pipe, by all means ship the box so that by default it reports these basic settings and indeed total packets tx/rx once every 48 hours to a central server, by all means, ship the box so it doesn't function at all until a unique admin password has been set.

I'd really like to log onto this box and see a webmin style graphical analysis of each IP address on my subnet and have that boken down into types of traffic, ooh, box #3 is using 45% of my traffic, and 66% of box #3 appears to be port 8080... hmmm

In an ideal world the 21st century ISP will also offer as an option a subsidised Locust Mesh box, with the market penetration an ISP has before you know it you have a cloud of mesh boxes providing a redundant and resilient alternative pipe, actually saving the ISP money, and if it was subsidised I really would not mind default parameters on it that ensured it worked best as a redundant backbone, priority to traffic controlled by the Smart Box.

Of course, what I am describing here has an analogy.

Virgin broadband and the other ISP's have in effect become MicroSoft.

The freaky stuff like smoothwall and locust and so on is Linux, it may be cool but stands zero chance against the MicroSoft in the business and financial real world.

But a Red Hat or Cobalt Raq or suchlike approach could work, the Locust product is *very* neat, the hardware is neat, the software is neat, the only thing missing is the business plan and implementation, there is nobody with any clout behind it.

But mainly, we need a 21st Century ISP, like Mark 14 years ago, he sold connectivity, period, what it said on the tin, period, with the best technology of the day to provide that and nothing else, period.

I'll leave you with a thought. As much as 10% of the UK electricity consumption is absorbed by telecomms networks, and while a redundant cloud of mesh boxes won't be able to compete on bandwidth, every single mesh node could be powered by solar PV and a lead acid battery for the nights.

So the future of computing is the return of the thin client, and that makes it all utterly dependent on the pipes being transparent and of sufficient capacity, and yet every single ISP out there, without exception, is working towards overstuffing the pipes and making them as opaque as possible.

If you thought malware and spam were the big wars on the network, just wait until traffic shaping vs port forwarding / tunneling / encryption starts to take off, you ain't seen nothin yet.

An ISP for the 21st century.
Tue Jul 22 07:23:49 -0700 2008
manage

It is not possible to avoid QoS.  They either impose it on your connection point, or a step or two up stream.

For example, if I put an ethernet switch in an apartment block and serve 20 apartments with 10 Mbps symmetric bandwidth, I'm allocating 2 Gbps to that group.  Assume I have a pair of GigE uplinks bonded together for 2 Gbps. All good so far.

Now, the apartment complex has 100 units which is 5 * 20 * 10, or 10 Gbps.  To assume no oversubscribption, I'd need a 10 GbE link just coming out of that apartment complex.  And if there are 100 such complexes in the city, I need a Tbps connection to meet me name plate commitment.  Now add in all the other customers, like businesses, etc. and the numbers get absurd quickly.

In the past, before P2P, this was easy to handle with oversubscription because most people don't use their connection to anywhere near capacity.  This is also the reason residential customers are usually forbidden from running servers, because it totally skews the model.  Want to run a server?  Pay for a business connection.

Oversubscription rates of 8:1 or 10:1 were good.  You'd never notice it, as long as everyone isn't running a P2P connection and consuming 100% of their allocated bandwidth.

QoS helps because it allows the system to flag packets with levels of importance.  No one is going to notice if an e-mail takes 2 seconds longer to get to its destination.  Nor does it matter if e-mail packets arrive at their destination out-of-order.  They all get reassembled before being delivered to your inbox.

VoIP, on the other hand, matters.  Get those packets out of order and you get choppy voice.  Delay the packets and you get dropped, or a choppy connection.  Quality suffers greatly.

Personally, I want an ISP who knows how to shape traffic based on traffic type and NOT source or destination.  I also want one who only does it WHEN NECESSARY and not just as a blanket.  Routers and switches should make decisions on throttling bandwidth based on total system resources available at that time, not just because it is a Bittorrent stream.

I also want an ISP that monitors traffic types and does what they can to internalize issues.  One that notices all that damn Microsoft update traffic and deals with MS to get an official "local" WSUS server that all traffic coming from his block to MS' home WSUS is redirected to.  One that sets up mirrors for updating Firefox, and possibly Debian, Fedora and the like depending on customer traffic.  A full mirror of the Linux .isos from various distros.

A transparent caching proxy can be a good thing.  How many people hit the BBC News or CNN or the Weather Channel on a regular basis?  During a "crisis", that would spike thru the roof.  Feeding everyone thru a properly configured Squid proxy would greatly lighten his load.

In short, I want an honest, competant, proactive ISP.

An ISP for the 21st century.
Tue Jul 22 08:31:08 -0700 2008
manage

It is not possible to avoid QoS.  They either impose it on your connection point, or a step or two up stream.

Who said anything about avoiding it?

A SmartBox as described will solve ALL the problems, want to run a P2p 24/7 and max out your nameplate bandwidth, you can, by altering the default SmartBox settings, and every 48 hours the SmartBox reports your custom config and actual throughput to the ISP.

At which point the ISP says...

"Charles, we have two options here, one is you reset the SmartBox to the defaults and let it throttle your P2p, two is you don't, at which point we bill you for a no-contention, not-shared, not throttled line, it's a 10 second config job at our end, choose one."

Problem sol-ved.

An ISP for the 21st century.
Tue Jul 22 08:06:28 -0700 2008
manage

The problem is if you're on the legacy v4 network you'll waste all your time avoiding spam and malware. Time for a superset in the same vein as usenet2 or bofh.*.

Then the device and service you speak of is practical.

I'll take two.

the war is well underway

Tue Jul 22 10:26:28 -0700 2008
manage

Yr post is so right there's little to do but elaborate a bit and minor corrections.

The war actually is underway, at least here in the US. Some of the channels I read are, literally, the "preview channel" for the next few days of IT-related tech news and the next several months of legislative and regulatory effort. Notice that I say "that I read" -- you have to be "a name" to post, mostly. But, the debate is well engaged. The FCC is holding hearings. DPI is turning into a political issue. It's fascinating and it's heating up.

"Nameplate" should actually be two sets of numbers: minimum guarantee and maximum theoretically possible (when there's no contention).

Someone above mentioned prioritizing VOIP. Nonsense -- or rather -- just what Guy says: do that, by consumer choice, at your own smart box. If reliable VOIP doesn't fit the nameplate minimum, then you aren't paying enough for connectivity or aren't using the right provider.

Guy mentioned the ultimatum to the guy who wants 100% P2P, 24/7. I don't think that that's quite right. Rather, actually, yes: protocol-neutral (IP-level, no DPI) throttling down to the nameplate minimum and not below. The actual amount of throttling depends on contention. Throttling should also favor "bursts" over "sustained". In other words, imagine two customers one of whom attempts to saturate, 24/7 and the other of whom only saturates now and again, 20 minutes at a time, those 20 minute bursts well separated in time -- give those 20minute guys priority, then degrade them if they keep going longer.

And no, Guy, the smart-box does not need to constantly report its configuration back to the ISP. There's no need for that as long as each consumer gets their minimum.

This will result in taking the gaming of the system to the next level, of course. If everyone did what I described then the next generation of P2P clients would would operate "round robin" in deliberate 20 minute bursts and, soon, Joe Casual user is again mostly frustrated. Well, so what? That game can keep escalating for a long time at low cost: Let Joe Casual make reservations; do real-time rate adjustment for usage over the guaranteed minimum; throw in lots of random params so that greedy P2P (or other) apps have to work really hard to hoard bandwidth; whatever.

Of course build more "pipe" as economics allow but, that gets to the root problem:

Early adopters had an experience similar to researchers who get huge budgets to build devices that, 5 or 10 years later, will be casual-consumer cheap. Researchers get spoiled by the capabilities that are fun to play with but reality is that "not everyone can have that" for a while to come. The reality is that that the ubiquitous net is going to be a lot slower than the net early consumer adopters found for a while, All of the really bandwidth-intensive (to the edge) business models are unrealistically ahead of their time in most parts of the world, even though there were quite a few years there when they could run just fine for an unsustainable small number of users. A lot of the political "debate" about DPI, QoS mgt. etc. is purely an attempt to tilt the regulatory field in favor of some business models and against others. P2P file sharing lacks a wealth of lobbyists. Commercialized media distribution has more than enough lobbyists. The criminal tainting of P2P follows and from that "So, of course DPI is reasonable. There's no other way to fight such crimes."

In reality, as the number of active users and uses goes up, we have to expect and embrace much lower bandwidth connectivity at the edge nodes -- to go backwards -- to build business models out of the reliable *minimum guarantees* rather than out of soaking up the overcapacity (relative to that minimum).

-t

the war is well underway
Tue Jul 22 11:20:37 -0700 2008
manage

And no, Guy, the smart-box does not need to constantly report its configuration back to the ISP. There's no need for that as long as each consumer gets their minimum.

No, the SmartBox does not HAVE to report back, you CAN disable it if you so choose, and the ISP simply drops you to a node with a higher contention ratio since you are in effect choosing not to play with your fellow node users.

The ISP absolutely HAS to have things going in its favour for this to work as a valid business model that is attractive.

Unless you are buying a dedicated leased line then your usage involves playing fair, if I'm on the same node as you there is no reason for you to subsidise my bandwidth.

The point you're REALLY missing about the SmartBox reporting back to the ISP is every user can see their own SmartBox data, so it forces both all the users AND the ISP to play fair.

SmartBox also automagically deals with all those users who simply cannot be bothered to ensure that their computers are not constantly spewing out malware, whether it is pwned mail bombed spam or simply the latest AVG no its not really a DDOS attack bit of coding... and while we are at it, FINALLY, a penalty for WIndows treating ALL packets as top priority.

close

Tue Jul 22 12:45:01 -0700 2008
manage

No, the SmartBox does not HAVE to report back, you CAN disable it if you so choose, and the ISP simply drops you to a node with a higher contention ratio since you are in effect choosing not to play with your fellow node users.

The smart box should let the user specify favorite applications (or, lower level, favorite protocols) with some parameters. E.g., maybe VOIP is critical from 0900-1700 but completely unimportant from 1900-0500. It should buffer and forward packets accordingly.

The smart box should measure actual demand for low-latency / higher-bandwidth and come up with demand curves for IP latency and bandwidth, weighted by the priorities of various protocols, minimum-acceptable demands, recent history, etc. But the curves themselves are just about raw demand for packet latency(to the next hop) and bandwidth -- nothing protocol-specific.

The smart-box can periodically send the ISP a spectral analysis of those demand curves. The ISPs job is to allocate its overcapacity such that the areas underneath the demand curves that are "filled in" by supplied overcapacity are equitable in the sense that all subscribers get their share of the sold overcapacity in proportion to the rate they're paying within some legislated maximum amount of deviation (or else refunds are due automatically, without court proceedings needed).

The ISP can still figure out from the reported demand curves a pretty good guess about each node's favorite protocols and general timing of uses of those. That's more than they'd get if they looked only at IP addresses and other transport protocol params. That's not a huge amount of data, compared to what ISP-level DPI gives them. Customers can also buy their way out of it with a refinement of this model:

Sure, by default, your reported demand curve has to be answered equitably relative to the rate you pay, within some legislatively minimal deviation.

If you like, though, you can have your smart box obfuscate your demand and report the obfuscated demand *along with* a measure of the obfuscation that increases the size of the deviation you are entitled to from an equitable distribution of overcapacity. You can tell the ISP less about your demand curves but, as a trade-off, your guaranteed share of the overcapacity drops.

There's no need for the smart-box to report its "configuration" to the ISP, optionally or not. Instead, just measure, weight, optionally obfuscate, and report those routing demand curves.

The ultimatum from the ISP comes if, for some reason, you insist that you are *not* obfuscating your demand curves (and thus are entitled to the minimal legal deviation from an equitable share) -- yet the ISP is unable to satisfy your demand, month after month, always owing you refunds while you insist you aren't obfuscating. Yet, somehow, the ISP doesn't have such problems with 99.99% of customers. Solution: it's incumbent on edge node owners to declare their effective level of obfuscation within regulated limits and if they can't meet that obligation then they either don't attach to the net or accept weakened guarantees (reduced refund conditions).

Control over the regulatory parameter "how much deviation from equity is ok" is very interesting in this sense: that parameter (if not 1) *forces* carriers to maintain *unused* over-capacity. As social policy, that is *precisely* what we would hope for -- it means we keep laying cable until we don't want any more. As economic policy, it is likely to *raise the price of bandwidth* until there's "enough" cable.

That just proves, by the way, that the mechanism proposed is correct. Of course the price of bandwidth has to go up if we want to replace today's false bandwidth claims with robust guarantees of bandwidth.

-t

the war is well underway
Tue Jul 22 12:45:02 -0700 2008
manage

Someone above mentioned prioritizing VOIP. Nonsense -- or rather -- just what Guy says: do that, by consumer choice, at your own smart box. If reliable VOIP doesn't fit the nameplate minimum, then you aren't paying enough for connectivity or aren't using the right provider.

That would be me, and you're wrong.  You're confusing bandwidth with latency.

A VoIP phone connection will only use about 24 Kbps peak per channel.  You can do VoIP on dialup, with the right codec.  The problem isn't bandwidth, it is latency.  Voice packets really must arrive at their destination not only in order, but within approximately 10 ms of each other.  Not a lot of leeway for waiting in line at the router.

Without QoS packets get processed at the router on a first come first serve basis.  If your neighbors are doing huge downloads, no matter what you set YOUR bandwidth for, you could easily have your phone call trashed by jitter as the ISPs router is queuing up packets.

With 802.1p QoS tagging you can tag packets as "latency sensitive" and get them to the head of the line for routing.  The beauty of it is most of the packets they are cutting in front of won't be affected.  We're talking milliseconds of difference in most cases and that just flat out doesn't matter to IM, e-mail, FTP or WWW traffic.

For this to properly work, 802.1p (or an equivalent) must be supported as far thru the network as possible.  Any ISP that offers VoIP probably supports it.  Any ISP worth their salt supports it.  My point is, it doesn't really help if only YOU on your Smart Box have QoS by the ISP and upstream equipment doesn't.

the war is well underway
Tue Jul 22 13:00:57 -0700 2008
manage

I was speaking imprecisely but I'm not confused in real life.

Right, demand actually two dimensions-- latency and bandwidth demands. At the application level you can characterize and weight those and come up with some two-dimensional measure of demand, over time. You can use user-declared weighting on which applications (hence protocols) are most important, when, and from that come up with the user's *declared* demand curve (er, surface). You can then (at the smart-box level) do DPI and *measure* demand, weighted by application preferences (the demand surface) and come up with an economically weighted aggregate demand surface to send to the ISP. Then regulate that ISPs have to spend their overcapacity equitably among those demand surfaces such that the shares given out are comparable to a rate-based fair share within some deviation.

It's perfect because regulators can slowly tighted the permitted deviation from equity (currently 100%) and, in so doing, *force* carriers to maintain some amount of un-used over-capacity -- which is exactly the right thing to be doing in the first place. All the economic incentives go in the right direction with that kind of regulation.

The only fudgey part is how to deal with users that fail to accurately report their demand curves and so the ISP, no matter how clever, can't deal them an equitable share of overcapacity.

-t

the war is well underway
Tue Jul 22 13:23:13 -0700 2008
manage

I was speaking imprecisely but I'm not confused in real life.

I figured as much, from the content of previous conversations.

Right, demand actually two dimensions-- latency and bandwidth demands.

Not really, in application.  The thing is there is no benefit to either the user nor the ISP for not using QoS.  The user's latency sensitive applications will perform sub-optimally, and the ISP's network will perform sub-optimally.  When QoS is on, everyone benefits.  With it off, everyone is penalized.  The entire demand curve is make-work.

Bandwidth, fine.  But there really isn't an over/under capacity in latency.  You can't "buy" more -- or less, as the case may be.

Keep in mind, I am talking basically QoS based on protocol for latency sensitive applications.  There IS a market for higher QoS for normal applications (like gaming), if someone wants to pay for it.  It depends on how the ISP is set up to handle traffic prioritization.

Usually, to do it right, VoIP traffic is shunted on its own VLAN and given tagged priority.  The VLAN itself is given a higher priority.

An ISP could create a "premium tier" and those customers could be moved wholesale into a higher priority VLAN for an extra fee.  A "gaming" tier.

In short, reinvent the ATM wheel.

the war is well underway
Tue Jul 22 14:06:15 -0700 2008
manage

Not really, in application. The thing is there is no benefit to either the user nor the ISP for not using QoS.

To be specific -- to the no doubt imperfect degree I grok the protocol parts that are relevant -- I'm saying that QoS is there, but quite abstract with a lot of control distributed to the edge:

The ISP has packet buffers which, idealized, are priority heaps. The comparison function used for those heaps is based on, sure, things like the 802.1p bits *and* which customer is transmitting that packet *and* that customer's declared "demand surface". The ISP empties the buffers according to a formula with no other inputs.

Bandwidth, fine. But there really isn't an over/under capacity in latency. You can't "buy" more -- or less, as the case may be.

Again, I'm not talking about regulating end-to-end latency. I'm talking about regulating latency just through the ISP -- how long a given packet is buffered. There can, separately, be latency promises between the ISP and his backbone. No one entity is liable for end-to-end latency, though. Latency through the ISP is reasonable to regulate just as it is reasonable to regulate latency through any one routing node. A regulatory demand for a single node is a requirement to maintain a certain level of capacity relative to what you sell of that node. It's like a reserve requirement in fractional reserve banking (thus, it *forces* the the network to expand (or slow down) when the alternative choice is to instead unilaterally break net neutrality).

There *is* over/under capacity of latency through a particular node. You *can* buy more or less (more or less tubes; more or less routing power).

-t

oh yeah, the other part

Tue Jul 22 13:28:16 -0700 2008
manage

For this to properly work, 802.1p (or an equivalent) must be supported as far thru the network as possible. Any ISP that offers VoIP probably supports it. Any ISP worth their salt supports it. My point is, it doesn't really help if only YOU on your Smart Box have QoS by the ISP and upstream equipment doesn't.

Sure, for end-to-end QoS. It's "end-to-backbone" that the ISP provides, though, and that's what I'm talking about regulating. I would imagine the right way to regulate backbones is just about the same, modulo edge:isp::isp:backbone.

Hmm. You could then still permit firms to be both backbone and ISP but they would be distinctly unable to favor their ISP service over others, even on their own backbone segments.

-t

oh yeah, the other part
Tue Jul 22 13:50:40 -0700 2008
manage

Most, if not all, backbones already do this.  It is really part of BGP and fairly common at the upper levels.  The big boys usually "do it right" amongst themselves.  It is the peons at the customer end that they don't care too much about.

My main issue is bastard ISPs like Comcast, who will give priority to THEIR VoIP and actually actively hobble competitors like Vonage.  My beef with big ISPs wanting to provide "QoS" is that they want to do it on a source AND destination basis, not protocol basis.

That is Yahoo or MSN could pay Comcast to bump all traffic on Comcast's network destined for Yahoo or MSN into the high priority lane.  Traffic to Google would get lower priority, etc.  I can see this for internal ONLY traffic, where traffic FROM a Comcast node TO a Comcast node would get higher priority, but NOT traffic destined for an exterior point.

This is the equivalent of turning the carpool lane on a highway into a "Ford" or "Nissan" lane.

oh yeah, the other part
Tue Jul 22 14:20:33 -0700 2008
manage

My main issue is bastard ISPs like Comcast, who will give priority to THEIR VoIP and actually actively hobble competitors like Vonage.

Right. So just say they can't do DPI and they must provided a metered level of equitable distribution across nodes they own relative to customer-supplied demand prioritizing declarations.

My best guess on evidence witnessed is that the massive push and deployment of that form of QoS has been largely a false front covering the installation of wiretapping hooks. Nobody in "middle" has any business doing DPI, at all, period.

-t

An ISP for the 21st century.
Tue Jul 22 11:44:11 -0700 2008
manage

The biggest mistake you are making is shopping for *residential* service. Shop for business ISP services and most offerings are much like what you are asking for - all provide (at least one) static ip, guaranteed bandwidth, etc. These days pricing here in the US for low end commercial services is between $40-$60 / month. Worth the premium, and the QoS works for you (rather than for them) since the bandwidth is guaranteed. For about twice what the low end costs you can get a much fatter guaranteed pipe and somewhere between 2-6 static ips.

The 'smart box' you will have to provide yourself, unfortunately. OpenWRT can be used to build a really feature rich firewall/router/DNS box.

An ISP for the 21st century.
Tue Jul 22 11:53:46 -0700 2008
manage

I'm in the process of shunting from Virgin domestic to NTL Business on the same fibre, site survey was today, changeover later this week hopefully.

And Open WRT will then become worthwhile, it wasn't on a strangled marketed to death bit of wet string.

An ISP for the 21st century.
Tue Jul 22 12:32:56 -0700 2008
manage

By a Smart Box*** I mean a box that performs the modem function, a box that povides DNS for my subnet, a box that provides a firewall, and a box that provides traffic shaping for my subnet, all in one, with a web based admin interface.

By "your subnet" are you referring to you getting multiple IPs personally, or your neighborhood where maybe 50 or so connections go to?

Caching DNS in the Smart Box, assuming the Smart Box is on your premesis, isn't of any real benefit.  Most ISPs run their own Caching DNS servers and as long as they are one or two hops away from you, that is fine.  The last thing an ISP is going to want is all that DNS traffic flowing from mini-DNS servers in the Smart Box.  Zero benefit to customer, more traffic on network.

That being said, I run my own caching DNS server on my home network.  But that is from a perspective of privacy, not network infrastructure.

As far as the rest goes, it is sort of a great idea.  It is great if the customer knows what they're doing.  It is a nightmate if Joe Schmoe has no clue and starts fiddling.

Make it an option for the technologically astute customers?  Well, they would be charging you for the box so why not just purchase your own and install OpenWRT on it?  It does all of the above plus more.

I have SpeakEasy DSL outside Chicago.  It costs me more ($105 / month) but I get exactly what I pay for (8 Mbps down / 1 Mbps up) and no throttling.  I can run in 24x7x365 and they don't care one whit.

In the house I have a Linksys WRT600N wireless router running OpenWRT.  I have a USB label printer plugged into it and shard.  I also have a Brother Laser plugged in via Ethernet and shared.  My VoIP DTA and a home server and jacked into 2 other LAN ports.  Five PCs, a Tivo and my OpenMoko phone talk via wireless.

I have QoS setup to give priority to VoIP and Teamspeak traffic.  E-mail, IM and HTTP packets are low priority.  Everything else is medium.  It works great.  That is the only way I can get everyone to work properly, no one complaining, no lag or dropouts.  All under my control.

An ISP for all things

Wed Jul 23 00:11:05 -0700 2008
manage

I’m sort of surprised no one mentioned the other reason many ISPs are keen to do traffic shaping of various sorts and it relevance to services Guy did not mention.  Most ISPs I know of want to sell you nearly everything under the sun.  They want to be your voice communication provider (via some sort of propriety VoIP).  They want to be your Video content provider (both ‘broadcast’ channels and video on demand / pay per view).  They want to be your audio content provider (via a menagerie of subscriber ‘channels’.  They want to provide you with local specific value added services such as local weather forecasts, local events, targeted community advertising.

 

My ISP has a dozen ‘zones’ of value added audio & video content of this nature. Pop Music (searching, lyrics, charts, performances), Radio, Porn, Sport packages, Team packages, Social sport content, Local live music coverage, local news and events, national news and events. Travel Planning and targeted advertising.  To me it’s no wonder that ISP with this business model are somewhat antagonistic to open or free purely internet based replacements for these services when they consume bandwidth the ISP would rather use for their business schemes.