funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
 ·Verizon Online DSL
 ·Skype
| reply to ib50MbSoon
Re: ISPs can't win!
said by ib50MbSoon  :If they don't throttle the p2p thieves, the 3rd party VoIP crowd won't be happy. Most members of the "3rd-party VOIP crowd" supports Network Neutrality. In fact, Comcast has announced a special collaboration with Vonage to ensure that its "protocol agnostic" solution doesn't affect Vonage. Comcast wouldn't have to do that if it behaved according to Internet Standards.
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
More fun, more features, Join BroadbandReports.com, it's free...
|
|
hottboiinnc
ME
join:2003-10-15
Cleveland, OH
 ·Time Warner Cable
·buckeye cable
| What internet standards? I didnt know there was a document that stated what every company had to follow. The rule is if you don't like what a company is doing or how you're treated you don't do business with them.
Simple as that. The same applies with Comcast- if you don't like the way they manage their network don't use it. |
|
bbenso1
join:2004-11-28
Baltimore, MD
| said by hottboiinnc :The same applies with Comcast- if you don't like the way they manage their network don't use it. But it's not always that easy. What if your choices are Comcast's network or no broadband? This is exactly the case in many places. |
|
espaeth
Digital Plumber
Premium,MVM
join:2001-04-21
Minneapolis, MN
·voip.ms
·Vitelity VOIP
·Callcentric
·VoiceStick
·ViaTalk
·Comcast
·Embarq
| reply to funchords
said by funchords :Most members of the "3rd-party VOIP crowd" supports Network Neutrality. In fact, Comcast has announced a special collaboration with Vonage to ensure that its "protocol agnostic" solution doesn't affect Vonage. Comcast wouldn't have to do that if it behaved according to Internet Standards. There are no standards that define the policies and criteria for how traffic must be differentially queued, only descriptions of the mechanics and theories about how such goals can be achieved. |
|
funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
·Verizon Online DSL
·Skype
| said by espaeth :There are no standards that define the policies and criteria for how traffic must be differentially queued, only descriptions of the mechanics and theories about how such goals can be achieved. That's right, and that's on purpose. The IETF seems to have structured this so that the end host should have the purview to set the handling instructions and that the network operator has the purview to decide whether and how to follow them.
That solves a number of problems, but it also creates a few in the application to Internet users that are increasingly non-technical in nature and many, many network programs that don't use DSCP since nobody has really done this on a residential network before.
I personally don't mind using opt-in DPI as a substitute for this gap. The DPI device could then mark the DSCPs according to signature, but do so acting on behalf of the end-user. I think that would solve a lot of problems that users are currently struggling with and it would help the ISP, too. Having it opt-in would hopefully drive the ISPs to act more honestly about what it does and it would require the user to actually be informed (versus some buried text in a TOS somewhere). Most users would want to turn this on.
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
More fun, more features, Join BroadbandReports.com, it's free...
|
|
jaminus
join:2004-10-14
Arlington, VA
| What's wrong with delaying non time-sensitive packets so time-sensitive packets can get to the next hop on a a timely basis during times of network congestion? Most people, myself included, aren't likely to notice an extra 100ms on an HTTP session. But that sort of jitter can severely degrade VoIP or online gaming. Inspecting each packet and classifying it by application type is one method ISPs can use to configure their systems to reduce the likelihood that time sensitive packets will suffer on account of network congestion. Application layer QoS really does make a lot of sense, even if existing standards don't spell out the details of a DPI-based traffic prioritization regime on the ISP level. |
|
funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
·Verizon Online DSL
·Skype
1 edit | said by jaminus :What's wrong with delaying non time-sensitive packets so time-sensitive packets can get to the next hop on a a timely basis during times of network congestion? Nothing.
said by jaminus :Inspecting each packet and classifying it by application type is one method ISPs can use to configure their systems to reduce the likelihood that time sensitive packets will suffer on account of network congestion. Except they can't always know whether a use is real-time or not. Identical algorithms in BitTorrent not only supports background file transfers, but it also supports real-time streaming. DPI equipment can only guess as to what it is.
said by jaminus : Application layer QoS really does make a lot of sense, even if existing standards don't spell out the details of a DPI-based traffic prioritization regime on the ISP level. It seems to be "the new black" but none of these many network vendor companies have made their case to the satisfaction of the IETF.
That said, I'm not against it myself -- but until it's Internet Standard, it needs to be done on an opt-in basis.
Secondly, it can't be an alternative to growing your network capacity as fast as you normally ought to -- and that's really what's behind this. For every dollar spent on DPI, ISPs are told that they'll save $10 on network upgrades. Now don't step in the marketing, but if that's even partially true, it's a Pandora's box leading to death by congestion.
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
More fun, more features, Join BroadbandReports.com, it's free...
|
|
hottboiinnc
ME
join:2003-10-15
Cleveland, OH | reply to bbenso1
I would use Comcast. I see no problem what they're doing. It's their network. follow the rules and be done with it. If you go over 350 or 400 gigs a month then you should have you account slowed or canceled. |
|
Lazlow
join:2006-08-07
Saint Louis, MO
| Hottboiinnc
Why? If all of an individual's downloads were done between say midnight and 8am, how is that hurting anyone else's service? This(effecting other people service) is not an issue of the amount that is downloaded or how it is being downloaded(for instance p2p) it is about when it is being downloaded(and of course greed). Downloading during those off hours also does not effect the ISPs capacity, so there is no extra capital expense for these downloads. The easiest way around this is to implement an open speed cap during off hours(to be automatically determined node by node on every X time frequency). Assuming that each node normally handles 30K users(without issue), when the traffic drops below 50% of capacity drop the speed caps. Let those that are downloading a lot (or anybody up at those hours) download as fast as the system can handle (say up to 90% load capacity of the system). That will keep the heavy users off the system during high load times (6pm to 11pm). If you could download after midnight at 50MB/s why waste your time downloading at 8MB/s during prime hours? Doing this would avoid caps (bad PR), throttling(more bad PR), and would cost very little to implement (always a good thing). The only thing preventing this model is greed. The ISPs see the HUGE markup for charging overages ($1.50 /GB that costs sub $0.12 /GB). But there is also a matter of the greed of users. Users have to learn that doing large downloads during prime time is inconsiderate to their fellow users (ie greedy). |
|
funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
·Verizon Online DSL
·Skype
| +100 points for Lazlow !
In fact, I think ISPs ought to take the top 5 most popular high-bandwidth background-transfer applications (4 or 5 of these will be P2P applications) and make specific voluntary recommendations for scheduling limits for upload and download windows. And as usage patterns change, they ought to revise them. Doing this would help both user and ISP.
For example, it might look something like this:
M-F 3:30 pm - 8:00 pm local ... 300 KB/s download limit, 11 KB/s upload limit
All other hours ... 500 MB/s download limit, 24 KB/s upload limit
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
More features, more fun, Join BroadbandReports.com, it's free...
|
|
Lazlow
join:2006-08-07
Saint Louis, MO
| I think in order for it to work that it would have to be dynamic (load based). Now how frequently the load would have to be checked and what percentage of total load capacity would be allowed is something that would have to be tested. As a first guess I would think every fifteen minutes would be sufficient with a 90% percent of capacity load cap. The load cap would insure that if somebody gets up at 2am to check the weather (or whatever) they would still see "normal" web speed.
Why such a low upload cap(24 KB/s)? Even Docsis 1.x has 9 Mbit/s usable (shared) bandwidth. The 90% (or whatever final number worked out) should take care of making certain there was sufficient bandwidth to insure "normal" web response. |
|
hottboiinnc
ME
join:2003-10-15
Cleveland, OH | reply to funchords
yah but then again you would have to resue Comcast and sue everyone else because they're telling you how you can use your connection. |
|
funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC | reply to Lazlow
I think I once figured out that 24 KB/s would not max out the system until 45-50 people were doing it -- it seems conservative, I know, but 25 people doing uploads would take up 4.8 Mbps of the available bandwidth. |
|
Lazlow
join:2006-08-07
Saint Louis, MO
| That is why my method is better. Assuming Docsis 1.x which has a shared usable bandwidth of 9Mbps. If you hold the cap at a level of 90% of 9Mbps(rough 8Mbps) then everyone can hold at a maximum upload rate with enough left over for people to still browse "normally". So if there was only one guy uploading he could have the entire 8Mbps, if two then 4Mbps, and so on. If you go up to Docsis 2.0 then the usable upstream goes to 27Mbps, 3.0 and it goes to 108. If the cap floats in this manner (assuming maintaining current tier floors) the maximum amount of downloading could occur while incurring virtually no additional expense to the ISP. The system would of course independently also handle downstream in the same manner with the downstream usable limits being 38 for both Docsis 1/x and 2.x, while 3.0 would be 152(4channel) and 304(8channel). Docsis 3.0 would also be limited in that most current home routers(internet side) and modems are only 100mbit. I still find it strange that a GigE router usually only has a 100mbit input. |
|
funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
·Verizon Online DSL
·Skype
| First off, I think that the ISPs ought to convey this information as advice for users to voluntarily follow, not only for the best experience for all users but also to be good enough for users who just want to "set it and forget it." It's just the right and smart thing to do. And if an ISP gets a speed-demon in its ranks, rather than reading him the riot act, maybe they can call and explain the problem, ask for his cooperation, and point to some ready-made pages that help him understand the problem and configure his system in a better way.
Secondly, I don't know that 90% of (anything) would leave a lot of room for "bursty" responses. At 90%, I think that most commercial gear are dropping packets preemptively anyway, trying to get the speedier senders to slow down (lookup Random Early Drop).
And finally, there is a reality that Cable wasn't built for uploading -- not in DOCSIS 1, 2, OR 3. Although the numbers have gone up through time, the asymmetry really hasn't changed all that much and its Cable's asymmetry -- a conscience design decision wholly within their control -- that is underlying their bandwidth "crisis."
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
More features, more fun, Join BroadbandReports.com, it's free...
|
|
Lazlow
join:2006-08-07
Saint Louis, MO
| The problem with set it and forget it, is that life does not work that way. Take weekdays from 9am-3pm during the school year, most systems are under a fairly light load. Now throw in a snow storm or any other event that closes school. Now suddenly the system is flooded at a time that it would normally be idling along. College usage is the same way. The week of and before exams everyone is hammering the books, the week after everyone is hammering the net. Life just does not follow a schedule.
The 90% was just a first guess. You would need a number that would always allow the stray individual checking the weather radar to be able to do so at a reasonable response time. But that "reserved overhead" (for lack of a better name), should be relatively small. Lets assume Docsis 1.0. If it is set at the 90% level then there would still be 4Mbs (38*.10) of bandwidth left over for our insomniac to check the weather. If he stays up longer (uses more bandwidth) the float would be cranked down (I think I suggested that it would be reevaluated every 15minutes). The 90% and 15 minute numbers would need to be dialed in to determine what fit the best.
I do not think that asymmetry is the source of the bandwidth crisis (I do not think there is a bandwidth crisis to start with). Most (certainly not all) people are more concerned with download speed anyway. While having no upload capacity to ask for the next whatever can be an issue it is a relatively rare thing. |
|
