Anxiously drumming your fingers while waiting for a Web page to download, you ask yourself "How much does my provider charge?!". If this scenario sounds familiar, you'll understand what drives research on "Quality of Service" on the Internet.
The problem is as simple as it is vexing: all Internet information is transmitted at the same rate, regardless of its urgency or importance. Consider an email message that won't be read until tomorrow, and a video being downloaded from a news site. Both compete on equal footing for precious network capacity, so the quality of service (QoS) deteriorates: the email delivery causes unnecessary "hiccups" in the video. Ideally, the network would give preferential treatment to high-priority data (such as video), transmitting less important information (such as email) at a more leisurely rate.
Currently, Internet service providers (ISPs) charge a flat rate, regardless of how you use your Internet connection. But ISPs could charge different rates depending on the capacity you require. Standard service might cost £20 per month, while £50 might guarantee hiccup-free video.
But how can ISPs ensure that such guarantees are met? One common strategy is based on "differential priorities". To understand it, you must first know a bit about how information is transmitted on the Internet.
Suppose you fetch a picture of a petunia from a Web server. The picture is too large to send all at once, so the server chops it into small pieces, called packets. The server transmits the packets to your computer, where they are reassembled.
Alas, when the network is busy, some packets never arrive. The network can handle only so many packets per second; when overloaded, the excess are simply discarded. Fortunately, your computer automatically determines which parts of the petunia are missing, and asks the server to resend them. (The server is even cleverer - it grabs as much as possible of the available capacity by sending packets faster and faster, slowing only when packets are lost.)
In today's Internet, packets are dropped arbitrarily. The Internet is essentially "fair": all transmissions get an equal portion of the available capacity. And therein lies the rub - the email/video example demonstrates that this simple notion of fairness is precisely what we are trying to avoid.
Back to differential priorities. To guarantee a specific QoS for certain kinds of information, the network first annotates each packet with its priority, like packages at the post office. Packets from email messages might be marked "whenever you get the chance", while video packets would exclaim "extremely urgent".
The packets are then transmitted as usual. If some must be jettisoned due to congestion, the network discards lower-priority packets first. Because the network can recover from missing packets, the email eventually gets delivered, but (as desired) the video takes precedence.
Differential priorities are a simple way to improve QoS, and piles of nasty mathematics can be used to define exactly what guarantees the ISP can safely promise. To oversimplify, as long as the ISP doesn't let too many people watch video at once, differential priorities guarantee hiccup-free viewing.
Economists looking at today's Internet observe that network capacity is inefficiently allocated. Everyone pays the same, so when an ISP increases capacity, users with modest capacity requirements subsidise the greedy pigs. Differential priorities are one way of rationally allocating scarce network capacity. While high-powered users might bemoan increased prices, ordinary users should see costs drop.
This capitalistic take on the Internet might make the socially-minded queasy. Should primary schools suffer poor QoS because they can't compete with investment banks? Certainly not - though if some users need subsidies, surely this policy decision should be explicit, rather than an unanticipated side effect of the Internet's inner workings.
Differential priorities are just one way of guaranteeing QoS. One alternative involves making reservations for network capacity, just like reserving a table in a fashionable restaurant. To guarantee a certain QoS, ISPs just need to refuse reservations when the capacity will be exhausted.
It's worth pointing out that some experts discount all this fuss over QoS. While today's networks are congested, network hardware is becoming very cheap and fast. Wait a few years, say the "throw capacity at the problem" crowd, and the problem will just evaporate. Moreover, today's network technology has the virtue of simplicity, while QoS proposals - both differential priority and reservation schemes - cause network maintenance nightmares.
When will ISPs allow you to purchase QoS guarantees? Most network hardware already support some sort QoS mechanism, but these features aren't widely used. One impediment to widespread acceptance is "backward compatibility": the QoS features must operate correctly even when communicating with sites that haven't adopted the same QoS mechanism.
Indeed, until QoS mechanisms are widely adopted, guarantees can be only as strong as the weakest link in the chain. If packets sent from Albuquerque to Cairo pass through a QoS-challenged ISP in Boston, then the transmission will suffer, no matter what QoS guarantees were purchased in Albuquerque. The Internet's oft-noted decentralised spirit means that video and netphone calls will suffer the occasional hiccup for some time. Nicholas Kushmerick is at: nick@compapp.dcu.ie For more information, see www.compapp.dcu.ie/nick/itr/