The Sprintlink network has changed dramatically over the last 18 months. This new architecture has taken extensive advantage of the nationwide Sprint SONET (four fiber) network.
----DS3-->router<-->FDDI<--->Router<----DS3---
Sprintlink makes use of backhaul to reduce their number of POPs.
More Evolution
All components are backed-up. If there is a break in the network, it will be rerouted in 30ms or less. The network is geographically diverse. Right now, the backup is in stand-by and can't be used. This may change.
Why not use ATM?
It's not as efficient as using IP over Sonet. The wide area circuits are the most expenseive and you want to maximize the use.
Based on data published by NLANR, the "cell tax" for typical IP traffic, over 20% of the bandwidth is lost to overhead.
There are a lot of advatages of IP over Sonet.
Louis Heatherton: Your peering point with other ISPs are horrible. What are you doing about it?
Ted says that he believe the NAPs have reached their limit. Louis says that these are private peering point. Vab says that private peering is done in DS-3 chunks. There is a plan to help address some of these issues; however, these issues are complicated by local loop issues as well as business and product management issues at both Sprint and the other peer.
Steve Swartz askes about QoS in the new Sprint architecture. What are you providing or plan to provide?
Vab understands the concern. They are working with their router vendor to implement what is possible. Ted says that implementing QoS on backbone routers is a problem at this time.
Michael Grave: How much demand have you seen for speeds greater than DS-3?
Ted says that folks are definitely wanting DS-3 and higher.
Michael Grave: One-to-One looks really good, How expensive is it?
Ted says that Sprintlink is just taking advantage of what Sprint did in building their SONET.
Michael Grave: What problems have been found?
The Bandwidth-delay product has been a challenge, but has been worked out now. It was important to get the buffer sizes right to insure no packet loss and rapid dranage of those buffers.
Mark Shapo: Are you planning to expand your customer access area?
Yes, as the 12000s get in the field, higher customer access will be more possible.
Cole from PSInet: Would you enuciate your peering policy?
Yes, please mail peering@sprint.net
Someone from UUNET Canada: How are you able to switch over from working to protect now?
Vab says that is must be done manually, but there is new capabilities being deployed in the next quarter that will fix that. Ted reinterates that a fiber cut in the SONET network is handled automatically now.
The same someone from UUNET Canada: I question the efficiency of using IP over Sonet. I recall doing TDM and how much wastage occurs.
Ted says that you really only have two choices: ATM and Packet over Sonet. The claim is that Packet of Sonet is more efficient than ATM.
Jay Adelson with the PAIX: Do you have any interest getting a private peering point?
Vab says that it is a product manager issue.
Scott Brasier: What happens when the backhoe problem has been corrected?
Ted says that it will be done manually.
What about giving customers some idea where the packets are flying?
Ted says that usually the shortest path.
Would customers be notified when such changes are done?
No, not usually.
Curtis notes that having an ATM backbone does not necessarily mean you can do QoS. This is because streams have to be classified into some various QoS. Having link encapsulation efficiency is not a complete solution to delivering that efficiency to the customer.