1 International IP Backbone of Taiwan Academic Networks Wen-Shui Chen & Yu-lin Chang APAN-TW APAN 2003 Academia Sinica Computing Center, Taiwan
Page 2 Current International Connections Japan – 155Mbps (Operated by ASCC) Japan R&E networks (APAN, SINet, KEK, Spring-8, CN-CERNet,...) NSPIXP-6 for IPv6 peering with Japan’s IPv6 ISP Hong Kong – 155Mbps (Operated by ASCC) Hong Kong (HARNet) US-West – 622Mbps (Operated by MOECC+ASCC+NCHC) PAIX for commercial ISP transit and peering Pacific Northwest GigaPoP for Internet-2 US-East – 622Mbps (Operated by ASCC) StarLight for next generation lambda-networks (Abilene, MREN, CERN, CA*Net4, ESNet, SURFnet, EU DataTAG, NORDUnet...) Europe-UK – 155Mbps (Operated by ASCC) DANTE/GEANT for UK & European R&E networks
Page 3 Current International Connections Japan US-West US-East Hong Kong HARNet Europe- UK 155Mbps 622Mbps
Page 4 From Users’ Point of View TANet Internet-1 (commercial) connection to U.S.- West (Seattle) The term of “User Communities” TANet2 Internet-2 (R&E) connection to U.S.-West (PNW Seattle) and STARTAP ASNet Asian connections –APAN networks (Tokyo, lots of ) –Hong Kong networks European connections –StarLight member networks (Chicago, lots of European nets)
Page 5 International Link Design Principle Research and education use first Fast recovery and self protection Minimize latency Minimize transit hops
Page 6 Research and education use first No bandwidth limit for R&E use Large margin for burst traffic Grid computing Virtual laboratory Earth observation Digital contents... Special configuration can be done on demand
Page 7 Fast recovery and self protection Multi-path design Diverse routes, backup each other Minimize path switchover overhead IGP (such as OSPF, is-is,...) may take tens seconds to re-compute new path Enable MPLS-Fast-Reroute to minimize the outage windows of path switchover End-to-end Self-healing Extend self-healing protection mechanism from the IPL-ADM to routers –APS 1+1 on SONET –MSP 1+1 on SDH
Page 8 Minimize latency Why? Standard TCP transfer rate is highly depends on end-to-end latency Virtual laboratory and real time works also sensitive to latency Shortest IPL path first route-once; switch-many Layer-3 routing table lookup will generate large latency –Reconstruct MPLS-LSP tables to make domestic and international backbone work together to avoid Layer-3 routing Layer-2 forwarding table lookup will generate a little latency –vender proprietary circuit-layer functions to avoid Layer-2 forwarding table lookup, for example Cisco’s Frame-Relay-Switching (on Packet over SONET/SDH line) Juniper’s Circuit-Connection and Translated-Circuit-Connection
Page 9 Recent project before the end of this year Taiwan to U.S. West coast Joint project of TANet, TANet-2, and ASNet Estimated bandwidth is 3.1 Gbps (5 * STM-4) Replace the exist Taipei-Seattle IPL (622 Mbps) Taiwan to Japan Upgrade the exist APAN TW-TokyoXP link (155 Mbps) New bandwidth is not fully confirmed yet Taiwan to Hong Kong Upgrade the exist TW-HARNet link (155 Mbps) New bandwidth is not fully confirmed yet
Page 10 Topology of TWAREN Backbone: 80 G GigaPoPs: 145 G Dark Fiber: 6 Taichung Hsinchu 20G 10G Tainan Taipei Sinica NCU NCTHU NCNU CCU NCKU NSYSU NTU NDHU NCTU NTHU 10GE 5GE fiber 10GE or fiber