7 ms·

Does anyone actually believe we will "run out of bandwidth" to the point where traffic shaping will become essential? Traffic shaping of saturated links sounds

Does anyone actually believe we will "run out of bandwidth" to the point where traffic shaping will become essential? Traffic shaping of saturated links sounds like a waste of time and resources when it's so much easier to just upgrade the hardware. It's cheaper too! It also happens to be the fucking purpose of the ISP (to provide adequate bandwidth for their customers). Think of how much it costs to pay people to manage an incredibly complicated traffic ruleset--not to mention the massively increased CPU overhead--versus a one-time hardware purchase. What you're talking about is either an ISP that's dying and can't afford to upgrade their shit or a greedy evil monster that simply wants to extract more profit from existing infrastructure in the most obtuse and invasive way possible.

woah 10y ago
Do you also argue strenuously against optimizing software CPU and memory consumption? > It's the FUCKING PURPOSE of AWS to run my code! Why don't they just buy more hardware to make it run faster?
- ardacinar 10y ago
  Has Amazon ever delivered less than what was advertised when you paid for it? I mean, I never experienced anything like that but geniunely curious if it is a thing (Might be possible in AWS lambda etc.)
  - oculusthrift 10y ago
    no. there is a service level agreement so if they don't meet the promises there is repurcussione. it's also not an apt analogy considering one can easily switch to Azure or Google Cloud or Self hosting whereas most people are stuck with one ISP that has a monopoly in that area
grandalf 10y ago
> Traffic shaping of saturated links sounds like a waste of time and resources when it's so much easier to just upgrade the hardware. It's cheaper too! It also happens to be the fucking purpose of the ISP (to provide adequate bandwidth for their customers). This is a very naive comment. For anything other than a dedicated "business class" circuit with hard performance guarantees, speculation and shaping is involved in delivering the end product.
- wtallis 10y ago
  > For anything other than a dedicated "business class" circuit with hard performance guarantees, speculation and shaping is involved in delivering the end product. The alternative to a connection with a SLA guarantee is a best effort connection, not discriminatory traffic shaping. Underprovisioning part of your network doesn't require you to show favoritism to certain traffic—the normal way to cope with larger than expected usage is to just start dropping packets where there's congestion. That doesn't imply or require traffic shaping.
  - grandalf 10y ago
    > just start dropping packets where there's congestion Aside from being normal, do you think this is preferable because it is fair? I think of QoS as a way of allowing more types of SLA to exist. Maybe Youtube makes a deal with T-Mobile to keep videos playing longer (before packets start dropping) in exchange for money. This money might allow T-Mobile to invest in other parts of its infrastructure. Isn't any backbone link in a sense a way of shaping traffic? The old pattern (from point A to B, say) may have been less efficient than the new pattern (using the new link). The link makes economic sense to install simply because there is demand for bandwidth from A to B. If there weren't, a link would have been installed elsewhere instead. The price of the new link is passed along to downstream and upstream providers. Assuming that each ISP has a shaping model in place that is generally fair, is there really any difference between these three things: - the ISP shopping upstream providers based on SLA committments for traffic X - the ISP imposing QoS that addresses the quality experienced by customers using traffic X - the ISP investigating upgrading networking hardware to allow for superior handling of traffic X I'd argue that in every case the ISP has a cost and a benefit. In the case of the QoS change, the cost is that some customers might be slightly worse off (those who used traffic Y but never used traffic X).
    - kelnos 10y ago
      > Maybe Youtube makes a deal with T-Mobile to keep videos playing longer (before packets start dropping) in exchange for money. This money might allow T-Mobile to invest in other parts of its infrastructure. That half of it isn't the problem, and I doubt anyone would argue against reasonable things that make it easier for a provider to increase their network capacity. The problem is what it does to YouTube's competitors: they now suffer because YT can afford to ask T-Mobile to make their traffic more important.
adrianpike 10y ago
> Does anyone actually believe we will "run out of bandwidth" to the point where traffic shaping will become essential? Traffic shaping has always been essential. Bandwidth has never been infinite, and there's always been latency vs. bandwidth tradeoffs that need to be managed.
- wtallis 10y ago
  > Bandwidth has never been infinite, and there's always been latency vs. bandwidth tradeoffs that need to be managed. Dropping packets has always been necessary. Traffic shaping has never been necessary, and modern fair AQM algorithms are a net win to both throughput and latency relative to the most commonly deployed congestion management strategies, so that latency vs bandwidth tradeoff is far more illusory than you think.
dv_dt 10y ago
As long as a competitive market exists, I'm happy to let the market work out who has the better strategy: hardware oriented vs rulesets, etc. But where there is no competition (e.g. the last residential mile), I want regulations protecting me as a consumer with various accessibility, privacy, and fairness guarantees.
nickbauman 10y ago
Unfortunately profit is almost always made from oversubscribing the lines.

woah 10y ago

Do you also argue strenuously against optimizing software CPU and memory consumption? > It's the FUCKING PURPOSE of AWS to run my code! Why don't they just buy more hardware to make it run faster?

ardacinar 10y ago
Has Amazon ever delivered less than what was advertised when you paid for it? I mean, I never experienced anything like that but geniunely curious if it is a thing (Might be possible in AWS lambda etc.)
- oculusthrift 10y ago
  no. there is a service level agreement so if they don't meet the promises there is repurcussione. it's also not an apt analogy considering one can easily switch to Azure or Google Cloud or Self hosting whereas most people are stuck with one ISP that has a monopoly in that area

grandalf 10y ago

> Traffic shaping of saturated links sounds like a waste of time and resources when it's so much easier to just upgrade the hardware. It's cheaper too! It also happens to be the fucking purpose of the ISP (to provide adequate bandwidth for their customers). This is a very naive comment. For anything other than a dedicated "business class" circuit with hard performance guarantees, speculation and shaping is involved in delivering the end product.

wtallis 10y ago
> For anything other than a dedicated "business class" circuit with hard performance guarantees, speculation and shaping is involved in delivering the end product. The alternative to a connection with a SLA guarantee is a best effort connection, not discriminatory traffic shaping. Underprovisioning part of your network doesn't require you to show favoritism to certain traffic—the normal way to cope with larger than expected usage is to just start dropping packets where there's congestion. That doesn't imply or require traffic shaping.
- grandalf 10y ago
  > just start dropping packets where there's congestion Aside from being normal, do you think this is preferable because it is fair? I think of QoS as a way of allowing more types of SLA to exist. Maybe Youtube makes a deal with T-Mobile to keep videos playing longer (before packets start dropping) in exchange for money. This money might allow T-Mobile to invest in other parts of its infrastructure. Isn't any backbone link in a sense a way of shaping traffic? The old pattern (from point A to B, say) may have been less efficient than the new pattern (using the new link). The link makes economic sense to install simply because there is demand for bandwidth from A to B. If there weren't, a link would have been installed elsewhere instead. The price of the new link is passed along to downstream and upstream providers. Assuming that each ISP has a shaping model in place that is generally fair, is there really any difference between these three things: - the ISP shopping upstream providers based on SLA committments for traffic X - the ISP imposing QoS that addresses the quality experienced by customers using traffic X - the ISP investigating upgrading networking hardware to allow for superior handling of traffic X I'd argue that in every case the ISP has a cost and a benefit. In the case of the QoS change, the cost is that some customers might be slightly worse off (those who used traffic Y but never used traffic X).
  - kelnos 10y ago
    > Maybe Youtube makes a deal with T-Mobile to keep videos playing longer (before packets start dropping) in exchange for money. This money might allow T-Mobile to invest in other parts of its infrastructure. That half of it isn't the problem, and I doubt anyone would argue against reasonable things that make it easier for a provider to increase their network capacity. The problem is what it does to YouTube's competitors: they now suffer because YT can afford to ask T-Mobile to make their traffic more important.

adrianpike 10y ago

> Does anyone actually believe we will "run out of bandwidth" to the point where traffic shaping will become essential? Traffic shaping has always been essential. Bandwidth has never been infinite, and there's always been latency vs. bandwidth tradeoffs that need to be managed.

wtallis 10y ago
> Bandwidth has never been infinite, and there's always been latency vs. bandwidth tradeoffs that need to be managed. Dropping packets has always been necessary. Traffic shaping has never been necessary, and modern fair AQM algorithms are a net win to both throughput and latency relative to the most commonly deployed congestion management strategies, so that latency vs bandwidth tradeoff is far more illusory than you think.

dv_dt 10y ago

As long as a competitive market exists, I'm happy to let the market work out who has the better strategy: hardware oriented vs rulesets, etc. But where there is no competition (e.g. the last residential mile), I want regulations protecting me as a consumer with various accessibility, privacy, and fairness guarantees.

nickbauman 10y ago

Unfortunately profit is almost always made from oversubscribing the lines.