Nuts & Bolts: A User’s Guide to ISP Network Management

by on February 24, 2009 · 16 comments

This is the third in a series of articles about Internet technologies. The first article was about web cookies. The second article explained the network neutrality debate. This article explains network management systems. The goal of this series is to provide a solid technical foundation for the policy debates that new technologies often trigger. No prior knowledge of the technologies involved is assumed.

There has been lots of talk on blogs recently about Cox Communications’ network management trial. Some see this as another nail in Network Neutrality’s coffin, while many users are just hoping for anything that will make their network connection faster.

As I explained previously, the Network Neutrality debate is best understood as a debate about how to best manage traffic on the Internet.

Those who advocate for network neutrality are actually advocating for legislation that would set strict rules for how ISPs manage traffic. They essentially want to re-classify ISPs as common carriers. Those on the other side of the debate believe that the government is unable to set rules for something that changes as rapidly as the Internet. They want ISPs to have complete freedom to experiment with different business models and believe that anything that approaches real discrimination will be swiftly dealt with by market forces.

But what both sides seem to ignore is that traffic must be managed. Even if every connection and router on the Internet is built to carry ten times the expected capacity, there will be occasional outages. It is foolish to believe that routers will never become overburdened–they already do. Current routers already have a system for prioritizing packets when they get overburdened; they just drop all packets received after their buffers are full. This system is fair, but it’s not optimized.

The network neutrality debate needs to shift to a debate on what should be prioritized and how. One way packets can be prioritized is by the type of data they’re carrying. Applications that require low latency would be prioritized and those that don’t require low latency would not be prioritized.


Cox’s Internet service, like most Cable internet services, was built on top of its cable TV service, which was designed to share TV signals in only one direction to households in a relatively small geographic area. Cable companies segment their networks into neighborhoods or “nodes,” with each node connected to a Cable Modem Termination System (CMTS). The size of each node can vary from a few thousand households to a few hundred thousand households. All cable Internet customers connected to a single node share the available bandwidth.

Here’s a simple analogy: Imagine you buy a house with your new spouse. The house has a tankless water heater that can provide an unlimited supply of hot water at a rate of 2-5 gallons per minute, which is adequate for the two of you. When you have houseguests, you manage the limited flow rate by having some people shower in the morning and some people shower in the evening. Then you have kids. As your kids grow up, you all need to shower around the same time in the morning and you experience hot water outages more and more often. You’re faced with two options: Continue to restrict how many people can shower at any one time, or buy a larger-capacity water heater. Substitute broadband for hot water and you’ve got the situation that ISPs are in.

As cable companies add more cable Internet subscribers and individual households use more bandwidth, the cable companies have essentially three options:

  • Segment their networks so each node is serving fewer users; or
  • Deploy new technology to increase the bandwidth of their CMTSes (e.g. DOCSIS 3.0);
  • Use the existing bandwidth more “efficiently.”

Using a network more efficiently means deploying some sort of “network management” system. Even though tankless water heaters can supply an endless amount of hot water, if you connect too many sinks and showers to a single heater and turn them all on at once, you will have a (temporary) hot water shortage. That’s why it’s usually not a good idea to run the dishwasher or washing machine when you’re taking a shower. Similarly, bandwidth on the Internet is only limited by the electricity needed to keep the routers running, but when everyone tries to use high-bandwidth applications (like streaming video) simultaneously, the network gets congested and slows down.

When thinking of hot water systems, washing machines and dishwashers can be thought of as non-time-sensitive uses of hot water because it’s usually not important when they’re done, as long as they’re done within a few hours of your preferred time. On the other hand, when you go to wash your hands, you want hot water immediately. This would be an extremely time-sensitive use. Showers probably fall somewhere in the middle. The same variety of time-sensitivity also applies to Internet applications.

When done right, network management is nothing to fear. It allows ISPs to provide better service to more customers at a lower cost. Hopefully, those customers will be happier because their time-sensitive applications will have enough bandwidth. And the lower costs to the ISP may result in lower prices to customers. For customers who want/need more bandwidth than average, ISPs can and do offer different levels of service.

Even in areas where the incumbent broadband ISP does not face any serious competition, network management is good for users: Without network management, it may be completely impossible on an overloaded network to make a VoIP call, remotely connect to your office network, or play online multi-player games.

Cox’s network management policy seems eminently reasonable. First, it only affects “upstream” traffic (i.e. traffic sent from users’ computers). The new system classifies all traffic as either “time-sensitive” (prioritized) or “less time-sensitive” (unprioritized). Unprioritized traffic includes FTP uploads, peer-to-peer file sharing, and Usenet posts. Most importantly, “Any traffic that is not specifically classified will be treated as time-sensitive.” Thus, the policy will not affect new Internet applications and anyone who encrypts their traffic (because using encryption prevents your ISP from being able to determine which application you’re using).

If you’ve noticed your Internet connection has suddenly slowed, your ISP’s new network management policy is probably not the cause. It may simply be that there are more households sharing the same last-mile connection and those households are using it more. What is needed are new metrics to compare broadband offerings. Heavy users of peer-to-peer file transfer applications may indeed see faster speeds by switching to an ISP that doesn’t use network management. But if all such users in a particular area switch to that ISP, the ISP’s network will likely quickly become overloaded and have to implement network management practices themselves. Just as insurance companies and financial institutions must avoid setting policies that attract the sickest or least-credit-worthy customers, ISPs may face the same problem of “adverse selection” by attracting the most bandwidth-intensive users if they do not either impose some form of network management or charge a premium for not limiting bandwidth.

New Metrics

Choosing an ISP based only on price and downstream rate is simply not enough anymore. The old adage that “you get what you pay for” still applies. The first thing bandwidth shoppers that have a choice between cable Internet service and some other form of Internet service like DSL or fiber need to realize is that only cable Internet services share the last-mile connection among multiple households. DSL and fiber services do not. Next, you need to understand that the quoted transfer rate is not guaranteed; it’s simply the fastest speed you can expect to obtain under ideal conditions–which may only occur when all your neighbors have their computers turned off). Beyond that, the following are some terms that should help you decide between ISPs and the different packages offered by each.

To return to the water heater analogy, if you move into an apartment building with a central tankless water heater, knowing the water heater’s flow rate is meaningless if you don’t know how many other people are living in the building and sharing the same water heater. Of course some people take longer showers than others. If how much hot water you get for your morning shower is really important to you, you may be better off finding an apartment with your own private water heater. But for those that will have to share a water heater with others, you’ll want to know the capacity of the water heater and the number of people it will be shared with.

  • Bandwidth – Bandwidth measurements are exactly like the flow rate measurement for tankless water heaters: It’s a measure of how much of some quantity (water or data) the system can deliver over a fixed period of time. Tankless water heaters are measured in gallons per minute. Bandwidth is measured in megabits per second. NOTE: Most telecommunications equipment measures quantities in bits (and multiples of bits such as kilobits, megabits, and gigabits) but most storage devices measure quantities in bytes (and kilobytes, megabytes, and gigabytes). When abbreviated, MB means megabyte and Mb means megabit. There are 8 bits in a byte, so a high-quality photo from a 6 megapixel camera (approximately 2.2 megabytes in size) would take about 3 seconds to transfer across an otherwise unused 6 megabit per second (Mbps) connection. For more about bandwidth and how it relates to latency, which is a truer measure of actual speed, refer to my earlier article in this series, “Some basic about edge caching, network management, & Net neutrality.”
  • Powerboost – This technology, now used by a number of ISPs, gives a speed boost to the first few megabytes of each upload and download. This is great for casual web surfing, but for large files the boost isn’t all that significant. With one ISP’s package, the speed boost is from 6Mbps to 15Mbps for only the first 10Mb of each download. This saves a maximum of 8 seconds per download regardless of how big the file is. When comparing packages, be sure to compare the actual download speeds as well as the boosted download speeds. In some cases, the actual download speeds are not reported in the ISPs advertising and you need to call to find them out.
  • Contention Ratio – This is the ratio of the total bandwidth promised to all users (based on their service plan) to the actual bandwidth available on the connection. If there are 2000 households, each with a 10Mbps plan, sharing a last-mile connection with a total capacity of 1Gbps, the contention ratio would be 20:1. To go back to water heaters: If each of 20 apartments in a single building is promised hot water at a flow rate of 3 gallons per minute, the building would need a heater with a flow rate of 60 gallons per minute to meet the demand if everyone takes a shower at the same time. That would result in a contention ration of 1:1. But if the building tries to save money by installing a cheaper heater with a flow rate of only 30 gallons per minute, the contention ratio would drop to 2:1. ISPs in the U.S. do not normally disclose contention ratios, but the practice is common in the U.K, where leading ISP BT has guidelines requiring a ratio between 20:1 and 50:1. There’s no way to determine your own contention ratio, but it might be worth asking the next time you’re shopping around for broadband service, if for no other reason than to raise awareness of this important metric.

In conclusion, there are a number of potential causes for a slow Internet connection and a number of possible solutions–but the deployment of network management systems by ISPs is probably not to blame. If anything, most users on such ISPs should notice their connections become faster for most applications. If you’ve ever had no hot water to wash your hands because someone was running the dishwasher, you’ll understand why network management is important. As long as an ISP isn’t using its network management system to favor one application over a competitor (e.g. prioritizing its own voice-over-IP (VoIP) service but not prioritizing other VoIP services), network neutrality advocates should have no cause for alarm. As explained above, Cox’s new system meets this test.

  • http://srynas.blogspot.com/ Steve R.

    Regretfully, I find many of the Network Neutrality posts to be carefully constructed to only present half the truth while purposely ignoring the deeper implications of what an unregulated packet flow could potentially mean. The simplistic focus is that managing a network is an engineering issue to the exclusion the concept of what a company may actually do to actively “manage” the flow of data.

    I do not have a problem with companies implementing “pure” engineering solutions to manage the flow of data. The issue (which is conveniently not disclosed) is that companies can manage traffic for undisclosed opaque “business purposes” that have nothing to do with engineering. There have been cases were telephone companies have blocked access to certain websites as an anti-competitive action or have degraded internet service as a means of “encouraging” customers to move to more profitable alternative services. Then we have the likes of the RIAA and the MPPA demanding that ISPs filter the data flow. The issue of net neutrality is not just limited to simple engineering, it must also consider how data flows can be manipulated to actively interfere with the free flow of data.

  • http://www.cs.princeton.edu/~tblee Tim Lee

    Adam,

    If I'm a Cox customer, why would I want Cox deciding which of my packets are “high priority” and which are “low priority?” If I want my BitTorrent packets to be high-priority, is there a way to request that, or are BitTorrent users forever second-class citizens? Also, what's to stop BitTorrent users from encrypting their headers to evade detection?

  • http://pobox.com/~adamm Adam Marcus

    Tim: You may not want Cox deciding which of your packets are high priority, but you may very well want Cox deciding that *your* VoIP packets (which are time-sensitive) should be prioritized over your *neighbor's* BitTorrent packets (which aren't as time-sensitive). There's nothing to stop BitTorrent users from encrypting their headers to evade detection (which would result in BitTorrent being prioritized). But if everyone (or, more accurately, every application) demanded prioritization, then prioritization is meaningless. Similarly, if every vehicles on the roads had sirens and flashing lights, then police, fire, and EMS vehicles would have *slower* response times. But that kind of prioritization is something that most everyone agrees is a good thing.

    A better solution to the network management issue may be for application developers themselves to decide whether their application needs to be prioritized and indicate such in the packet headers, but to my knowledge there isn't a network management system in use that would allow for that. There is also the concern that if application developers could do this with no consequences, then every application developer would do so and we'd be no better than if we didn't have network management.

  • Ryan Radia

    What are your thoughts on DiffServ (http://en.wikipedia.org/wiki/Differentiated_ser… I think the idea would be that ISPs maintain not just bandwidth tiers but also priority tiers, such that pricing would differ depending on the priority level of a packet.

  • http://pobox.com/~adamm Adam Marcus

    I think DiffServ is a great idea, but wasn't really designed to be used by end-users. And my understanding is that there really aren't any mechanisms in place to deal with packet prioritization across networks owned by different companies. But Bandwidth Brokers (http://en.wikipedia.org/wiki/Bandwidth_Broker) may provide a solution for both end-users and inter-ISP networks.

  • http://www.cs.princeton.edu/~tblee Tim Lee

    But if everyone (or, more accurately, every application) demanded prioritization, then prioritization is meaningless.

    Exactly. Which is why prioritization based on application isn't going to be a viable long-term strategy. No application developer is going to voluntarily downgrade the performance of his own software. Nor is a user going to meekly accept slower performance merely because an ISP has decreed that he's running a “low-priority” application. This is why any viable network management strategy needs to let end users, rather than network owners, decide which packets are high-priority, and to do so in a way that gives them an incentive to be sparing with the “high priority” flag. In other words, any practical prioritization scheme is likely to be consistent with the end-to-end principle.

    Personally, I have my doubts about whether prioritization is useful at all, but if it is made to work, it will be an end-to-end-friendly protocol like DiffServ, not a clumsy hack like Sandvine.

  • Ryan Radia

    What are your thoughts on DiffServ (http://en.wikipedia.org/wiki/Differentiated_ser… I think the idea would be that ISPs maintain not just bandwidth tiers but also priority tiers, such that pricing would differ depending on the priority level of a packet.

  • http://pobox.com/~adamm Adam Marcus

    I think DiffServ is a great idea, but wasn't really designed to be used by end-users. And my understanding is that there really aren't any mechanisms in place to deal with packet prioritization across networks owned by different companies. But Bandwidth Brokers (http://en.wikipedia.org/wiki/Bandwidth_Broker) may provide a solution for both end-users and inter-ISP networks.

  • http://www.cs.princeton.edu/~tblee Tim Lee

    But if everyone (or, more accurately, every application) demanded prioritization, then prioritization is meaningless.

    Exactly. Which is why prioritization based on application isn't going to be a viable long-term strategy. No application developer is going to voluntarily downgrade the performance of his own software. Nor is a user going to meekly accept slower performance merely because an ISP has decreed that he's running a “low-priority” application. This is why any viable network management strategy needs to let end users, rather than network owners, decide which packets are high-priority, and to do so in a way that gives them an incentive to be sparing with the “high priority” flag. In other words, any practical prioritization scheme is likely to be consistent with the end-to-end principle.

    Personally, I have my doubts about whether prioritization is useful at all, but if it is made to work, it will be an end-to-end-friendly protocol like DiffServ, not a clumsy hack like Sandvine.

  • http://www.laptopbatteryclub.com/ laptop battery

    This is great news. Best of luck for the future and keep up the good work.

  • http://www.laptopbatteryclub.com/ laptop battery

    This is great news. Best of luck for the future and keep up the good work.

  • Pingback: site link

  • Pingback: Total Hair Regrowth Review

  • Pingback: DDOS Protected Webhosting

  • Pingback: yahoo premier league

  • Pingback: natural health products

Previous post:

Next post: