Network Computingwww.networkcomputing.com

Private Peering: Nothing But Net
The gathering places of the packet powwows in years past were at the various NAPs (Network Access Points). Nationwide providers struggled to reinforce these pipes, while local providers struggled just to get there. The trends have changed, however. After living through too many days of high packet loss, route flapping and just downright bad weather at these public exchange points, architects have devised a new approach: Private peering forges dedicated connections between large providers to aid general data flow. Private peering not only alleviates some of the rush-hour traffic at NAPs, it also promotes fast and efficient connectivity between backbones.

Do private peering agreements and routing policies make that much of a difference? To answer this question we did some testing of our own. Using three geographically separate sites, we ran a series of FTP transfers for several days. Sites A and B were in the United States on different backbones, while site C was in Europe. All were attached at fractional T1 speeds or greater. To keep the test accurate, line utilization and server loads were monitored.

File transfers from site A to site B averaged 22 KB per second, peaking at about 56 KB per second. File transfers between site B and site C averaged between

11 KB and 12 KB per second. Yet transfers from site A to site C crawled at no greater than 3 KB per second--the speed one would expect using a 28.8-Kbps modem. In an attempt to get to the bottom of this discrepancy, we began using the trace route utility. We plotted the paths among all three sites and compared routes. Site A was on MCI's backbone, while site B resided on ANS' network. The private exchange point between the two providers made our sites physically closer, which helped account for the increase in speed. Connectivity to site C, however, relied on the NAP in Chicago before switching over to the European provider. It soon became obvious that peering has definite advantages.

Although it's almost impossible to say for sure whether this is the sole factor in our poor performance between sites A and C, similar tests reinforced this observation. Two things should be noted here. First, providers that continue to strive to be well-connected using private peering will most likely demonstrate consistently better performance. Second, anyone wishing to use VPNs for branch offices should try to stay on the same backbone. If this isn't possible, you should at least look into the routing agreements between providers before making any commitments.

Enterprise ISP Real-World Pricing
In the past, attempts to provide accurate pricing scenarios have ended up as nothing more than reprints of whatever the ISP marketing teams throw out. In an attempt to obtain more accurate information, we decided to borrow from our RFP (Request for Proposal) model and create two mock companies. We provided location, area code and telephone prefixes along with our requirements to the ISP sales teams. We then had them price out an actual quote. The goal of this exercise is twofold: one to provide real-world pricing, and two, to demonstrate how availability (point of presence distribution) and local loop costs can heavily influence total cost.

Company A is a midsized organization looking for full T1 service for its headquarters. The majority of Internet-related services, including Web and e-mail, will reside in-house, but managers would like the provider to take care of its DNS services. Company A is look-

ing at signing a one-year contract, but it's also interested in any pricing breaks for a three-year commitment. Its main location is in suburban Chicago.

Company B is a smaller organization with a less immediate need for bandwidth. Rather than pay for a larger, dedicated pipe to host its Web server, it's looking to off-load the hosting of its site in return for quicker Web service and lower line charges. Asking for an initial fractional T1 (128-Kbps line), company B is willing to sign only a one-year agreement. It's located in the heart of Chicago.

Note that with some providers, the local loop-pricing differences between suburban and downtown is moot. With others, there's a much greater discrepancy. In many cases multiple POPs in a metropolitan area can play a big role in total cost. In turn, if an organization is in a rural area, the ISP decision could be heavily influenced simply by local availability.