State of the Internet:NSPs: ANS

Curtis Villamizer

Table of Contents

• Slides from this talk
• Stan Barber's Notes
• May 1996 NANOG Home Page

Stan Barber's Notes

Two Networks

AS690 NSS based backbone

AS1673Bay Networks BCN based backbone

ANS Cisco 7000/7500 based concentrator AS

• End customer attachments here

AS690 NSS based backbone

• 
• RS6000/930 with AIX and gated for management and routing
• Custom FDDI and HSSI interfaces for fowarding

AS1673 BayNetworks BCN based backbone

• Same cities as AS690 -- more DS3 capacity
• Dual 604/100 Risc CPU for management and routing
• 68040 based HSSI and FDDI forwarding cards
• Dual 604 OC/3 forwarding cards (considering)

ANS Cisco 7000/7500

• Multiple concentrators at backbone POPs (3 or 4 FDDIs at each POP)
• Each POP has its own AS for IBGP scaling
• Route reflectors are used

Performance issues

• NSS routers are stable, but limited PPS
• Cisco 7500 and BayNetworks BxN are much less PPS limited, but neither can handle a full DS-3 with significant route flap
• Performance criteria for new backbone routers
  • routing is stable under extreme adverse conditions
  • no packet loss to stable destinations
  • performance degrades gracefully under mild congestion
• Other considerations
  • control over routing policy
  • dynamic reconfiguration
  • secure operation

Problems/Issues

• Growth path beyond DS3 -- currently just adding more DS3
  • ATM at OC3c? ATM VCs at > 45Mb < OC3c? IP/Sonet?
  • What can be purchased?
  • Router vs. switched solution?
• Fallback with circuits fail
  • Sufficient backup capacity
  • graceful degradation if congestion occurs
  • Weighted Fair Queueing?
• Failure to gracefully handle intermitents
  • failure to detect (probe packets inadequate)
  • switching back and forth (no damping)
  • solutions are LQM and IGP link state damping
• Scaling of global routing
  • ANS need to reliably route our own customers
    • need accurate orgin AS with prefixes
    • in the absense of au-nums, need policy on per origin AS
    • investigating infering policy from observed AS paths
  • need accurate information in order to aggregate [Europe and Canada seem to have good registry info.]
  • backend to automate the configuration is in place
  • our own IBGP scaling is under control due to use of multiple AS and route reflectors within AS

Curtis notes that there have been problems with loss at the public interconnects due to problems that were probably largely unrelated to the interconnects themselves.