Network Computingwww.networkcomputing.com

As the Internet's focus shifted to business, ISPs proliferated, and customers demanded better service. ISPs answered their customers with the SLA (service-level agreement)--a series of promises for maintaining a satisfactory level of connectivity between the ISP and the customer. Sadly, though, these documents are hardly worth the paper they're written on and surely won't compensate you fairly for the business you lose when your site goes down. After all, while compensation for an SLA violation is nice, it won't offset the ill will and loss of business that outages cause, and thus needs to serve as a deterrent to the ISP. The remuneration must be sufficiently high to affect the ISP's business.

We asked 16 nationwide ISPs to tell us all about the carrots they dangle in front of customers. Of those 16, seven were willing to come to the table: Cable & Wireless Communications, Concentric Network Corp., Epoch Internet, Qwest Communications International, Savvis Communications Corp., Sprint and UUNet. Of the rest, Excite@Home declined because it offers no SLAs to its residential customer base, and the remaining companies--GTE Internetworking, IBM Global Services, IDT Corp., IXC Communications, Level 3 Communications, PSINet, Verio and Vnet--didn't respond to our queries. The seven participants provide SLAs to their customers and offer similar guarantees, but Epoch's promises stood out as the best--that is, the least bad. Individual details of various SLAs might appear better than Epoch's, but Epoch backed up its agreement with heftier customer credits for outages and latency violations than those provided by any other company.

Dangling Carrots

ISPs generally offer latency and outage guarantees in their SLAs; a few even issue a third guarantee against packet loss. Latency guarantees are meant to assure customers that data will travel from the entry to the exit of the ISP's network with minimal delays. The highest acceptable latency in our survey was 85 ms, guaranteed by Epoch and UUNet. While that sounds reasonable at first, all the ISPs use monthly averages instead of individual latencies. Therefore, there's no guarantee that any particular packet will arrive speedily. Some data may slog across the network at 800 ms, while other packets breeze by at 20 ms. If the ISP can't meet its guaranteed average latency, no credit is enough: The company should offer to find you a better ISP.

Outage guarantees aren't much better. They're designed to ensure that a customer's connection to the ISP's backbone will be maintained for a given amount of time every month. The ISPs in our survey promise 99 percent or better availability. But the amount of business lost in even a one-hour outage within a month (meaning the network is available about 99.98 percent of the time) can be significantly greater than the remuneration the ISPs offer. Outage guarantees range from a minimum of one day's credit for an hour of outage to an entire month's credit for more than eight hours of downtime, as offered by Epoch. Sprint and UUNet also offer up to a month's credit, but that credit doesn't kick in until service has been down for more than one full day.

The third type of guarantee, offered by Cable & Wireless, Concentric and Savvis, provides credit if a certain level of packet loss occurs on the host network. This guarantee is the most difficult to monitor, since the network on which the packets were dropped is almost impossible for a customer to track. Several conditions could cause packet loss, and most are usually out of the ISP's control. A significant packet loss in a short time typically indicates a problem on the line--perhaps with the last-mile connection, if not somewhere else. Therefore, when such loss occurs, it's a good idea to open a trouble ticket with the ISP or last-mile provider.

Most of the credits ISPs promise are linear--a mere one-day credit for every hour without service. But linear agreements don't acknowledge that outages have a cumulative effect. The longer an outage lasts, the more damage is done, and we think the subsequent credits ISPs promise should multiply exponentially.

An SLA should be written to show customers that a provider is willing to be fiscally responsible for its outages, and will compensate customers fairly if it doesn't hold up its end of the bargain. If a customer hurts, so should its ISP.

Only If You Ask

What if you do indeed deserve a credit? ISPs have three basic policies for granting credits for SLA violations: automatic, by request and by logging a trouble ticket. Epoch, Sprint and UUNet all require customers to request a credit; that is, the customer must call or send e-mail to inform the ISP that a violation occurred and ask for credit. Although all the ISPs but Cable & Wireless monitor their circuits and can tell when an outage occurs, only Concentric provides automatic credits--a feature we believe should be part of every SLA.

Cable & Wireless, Qwest and Savvis weigh down the opposite end of the benevolence scale. Not only is credit not automatic, but these three ISPs require their customers to have logged trouble tickets during any outages to get credit. Talk about adding insult to injury. First you have to call the ISP's NOC (network operations center) to log a trouble ticket during the violation, and only after the network has been repaired can you call to request a credit. Customers should not have to provide the effort needed to monitor their Internet circuits to receive credits.

The Winner

Fishing a winner out of this sea of sorry SLAs was an unpleasant task. In our survey, we asked specific questions about what the companies offered, and how customers would be remunerated.

We also wanted to know how the ISPs routed traffic, and how much of their networks were utilized on average. We asked these questions because ISPs can guarantee a latency level only for data traveling on their own networks. Once the data is handed off to the destination host's backbone, all bets are off.

Packets can be routed in two ways, known as "hot potato" and "cold potato." Think of the childhood game, and you'll get the idea. In hot-potato routing, if a customer's packet is destined for another ISP's network, the customer's ISP tries to hand it off to the destination ISP as quickly as possible. Cold-potato routing, on the other hand, means an ISP holds on to customers' traffic and tries to get it as close as possible to the final destination before delivering it to another network. We gave higher marks to the ISPs that use cold-potato routing because that method keeps data better covered by the SLA.

To choose our winner, we focused most heavily on how the ISPs handle traffic and how they credit their customers. We also placed importance on the amounts credited--whether they were linear or were ramped up as violation lengths increased. After all, what good is an SLA if the ISP offers only token compensation?

Having completed our calculations, we reluctantly chose Epoch's SLA. True, the company doesn't offer automatic credits. Its latency guarantees are also considerably worse than most. But Epoch does the best job of awarding credits to customers that it has let down. Those credits multiply quickly as the lengths of outages increase, demonstrating that this ISP is committed to making sure its customers are happy--or at least compensating those customers decently when they're not.