.--.        adam@ndtwin
    |o_o |       -----------
    |:_/ |       Host       NCKU CSIE × Purdue ECE (dual degree candidate)
   //   \ \      Role       Research Assistant @ Academia Sinica NSL
  (|     | )     Focus      NDTwin, P4/BMv2 development line
 /'\_   _/`\     Languages  C · C++ · Python · P4 · Java · Verilog · MATLAB · asm
 \___)=(___/     Research   learned indexes for packet classification,
                            measurement validity in BMv2 benchmarking
                 Contact    adam010341@gmail.com

• NCKU CSIE, Junior · Purdue ECE–NCKU CSIE Dual Degree Program candidate
• GPA 3.85/4.0
• TOEIC 980/990

• Summer intern, Network and System Laboratory, Academia Sinica — Jul 2026 — Sep 2026
• Research Assistant, Network and System Laboratory, Academia Sinica — Sep 2026–
• Undergraduate Researcher, Computer & Internet Architecture Lab, NCKU — Jul 2026–
• Class representative, Purdue ECE–NCKU CSIE Dual Degree Program

• Learned index structures for packet classification (capstone, NCKU CIAL, advised by Prof. Yen-Kuang Chang). Patches and an exhaustive certifier: nuevomatch-error-bound-fixes
• Measurement validity in BMv2 benchmarking (preprint in prep) — Independent research

NDTwin keeps a live copy of a network: its topology, every flow and the bandwidth it uses, and which switches are on. Apps try a change on that copy first, then push the best one to the real switches.

What it is used for:

• Some links are congested while others sit idle. The traffic-engineering app watches every ECMP group and re-maps flows so the group's links share the load.
• The network is nearly empty at night, yet every switch stays powered. The energy-saving app turns off switches whose links fall below a low watermark, and turns them back on when traffic returns, without lowering QoS.
• "What if I turn these switches off?" Apps simulate many what-if cases in parallel on the twin's current state, pick the best outcome, and apply it in real time.
• Writing your own network app. The kernel's REST API covers topology, flows, rules, power and fault injection. A traffic generator launches thousands of TCP/UDP flows from one config file, and a Mininet network lets you test before going to hardware.

NDTwin runs on OpenFlow hardware switches or on Mininet. On a P4 switch, its sFlow scheme (IEEE ICC 2026) can sample as often as every packet with zero control-plane overhead, so even small flows are detected and measured accurately.

My part: a P4/BMv2 backend next to the original Open vSwitch + Ryu one, behind the same kernel API, so behaviour that lives in a P4 pipeline can be studied with the same twin. Apps that rely on OpenFlow group tables, such as traffic engineering, stay on OpenFlow.

Code: ndtwin-lab (kernel, apps and tools) · NDTwin-Kernel-P4 (P4 backend, testing release)