------ =_NextPart_000_01BD7D5E.B034C7AE
Content-Type: text/plain
InterNAP is a exchange point provider that peers with 8 of the big
backbones. They are centered here in Seattle, but they just opened a
NAP in NYC as well I think (SFO, Chicago, and Atlanta to follow this
year). Anyway, this page talks about something they have called
"assimilator" that is supposed to pick superior routes. Anyone know
anything about this?
David, didn't you do a case study on InterNAp for your bus school class?
Do you know anyone over there we could talk to?
- Stefan
<<c02snet.htm>>
------ =_NextPart_000_01BD7D5E.B034C7AE
Content-Type: application/octet-stream;
name="c02snet.htm"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
filename="c02snet.htm"
<BASE HREF=3D"http://www.nelsonpub.com/cn/c02snet.htm">=0A=
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">=0A=
=0A=
=0A=
by Jennifer Lind, Special to Communications News
=0A= =0A=Seanet—one of the largest local ISPs serving northwest=0A= Washington state—faced an enviable problem early last year:=0A= demand for services was approaching current bandwidth capacity. A=0A= sophisticated application of proprietary TCP/IP routing=0A= technology not only provided capacity on a scaleable basis, but=0A= improved connectivity as well.
=0A= =0A=Seanet had single T1 connections to Sprint, MCI, and Agis. The=0A= three connections provided redundancy, as well as the ability to=0A= police traffic and balance routing among the different backbone=0A= providers.
=0A= =0A=The three-T1 infrastructure allowed Seanet to route traffic to=0A= addresses within Sprint's backbone via Sprint, MCI addresses via=0A= MCI's backbone and Agis addresses via Agis. For addresses that=0A= did not reside on these three backbones, traffic was routed=0A= indirectly via the optimum default route for Seanet, MCI. This=0A= meant that all default traffic as well as MCI traffic routed via=0A= MCI, often resulting in an overload of the MCI T1 and slower=0A= performance for Seanet customers. Furthermore, without a metered=0A= billing system, Seanet still paid for unused bandwidth on=0A= connections that were not at capacity.
=0A= =0A=Seanet had several options to increase and better manage=0A= bandwidth. One was to buy an additional T1 to a fourth backbone=0A= provider. This option would add capacity but might do nothing to=0A= relieve congestion on the MCI connection; and if Sprint became=0A= the congested route in six months, a second MCI T1 would do=0A= nothing to relieve the Sprint congestion.
=0A= =0A=Another option was a single T3 to one provider. This would add=0A= capacity and eliminate the need for routing. It would also,=0A= however, eliminate redundancy: If that provider's system went=0A= down, so would Seanet.
=0A= =0A=Seanet needed more capacity with direct connections to the=0A= major backbone providers for better traffic routing at an=0A= affordable price.
=0A= =0A=The solution was a DS3 to InterNAP Network Services, which=0A= provides direct connectivity and routing management with a=0A= metered billing plan comparable to the cost of several T1=0A= connections. InterNAP's private network access point (P-NAP )=0A= provides high-bandwidth direct private peering between Seanet=0A= customers and the seven major backbones.
=0A= =0A=However, a seven-time multi-homed infrastructure alone offers=0A= no significant advantage over a single connection other than=0A= redundancy, unless traffic is routed to the backbone that owns=0A= the destination address.
=0A= =0A=To combat random, inefficient routing, InterNAP uses a=0A= proprietary, patent-pending technology that directs routers=0A= throughout the Internet to choose the fastest and most direct=0A= path to and from InterNAP. This=0A= technology—Assimilator—allows the P-NAP to determine=0A= which routes are connected to which backbones and to place all=0A= Seanet customer outbound traffic directly onto the destination=0A= backbone in one router hop. Likewise, inbound traffic destined=0A= for Seanet customers travels only over one backbone, the fastest=0A= and most direct path possible, to the P-NAP, and to the = customer.
=0A= =0A=In effect, the P-NAP treats each backbone to which it connects=0A= as a private network and only sends and receives IP packets=0A= destined for addresses directly connected to that network. The=0A= ability to keep outbound and inbound network traffic on a single=0A= backbone from source to destination and destination to source is=0A= called symmetric routing.
=0A= =0A=This proprietary routing technology bypasses 80% to 85% of the=0A= major Internet congestion areas, the public network access points=0A= and exchanges (MAE-East, MAE-West, etc.) where data packets are=0A= sorted and swapped among backbone providers. These public=0A= exchange choke points are the major cause of slow performance and=0A= download latency for the mass audience of the Web. The packet=0A= loss at these points is generally agreed by Internet engineers to=0A= range from 10% to 40%, but without centralized management there=0A= is no official count.
=0A= =0A=The P-NAP allows Seanet customers to circumvent the choke=0A= points, eliminating packet loss and improving the speed of=0A= connectivity. And with a DS3, capacity is highly scaleable,=0A= permitting Seanet to accommodate more customers.
=0A= =0A=Between an ISP and its market audience, the quality and=0A= reliability of Internet connectivity will increasingly become a=0A= matter of brand protection—America Online's settlement with=0A= disgruntled customers is a prime example—and this will=0A= further fuel the demand for better performing, faster Internet=0A= connections.
=0A= =0A=But the technology applies not just to ISPs, but to enterprise=0A= companies looking to facilitate traffic to their Web sites (The=0A= Seattle Times), improve online gaming performance (RTIME, Inc.)=0A= or speed metasearch queries (go2net's MetaCrawler). InterNAP's=0A= technology is in use by more than 50 clients, including=0A= Microsoft, Adobe, and Amazon.com.
=0A= =0A=
------ =_NextPart_000_01BD7D5E.B034C7AE--