pith. sign in

arxiv: 2604.14650 · v2 · submitted 2026-04-16 · 💻 cs.DS · cs.NI

PlanB: Efficient Software IPv6 Lookup with Linearized B^+-Tree

Zhihao Zhang , Lanzheng Liu , Chen Chen , Huiba Li , Jiwu Shu , Windsor Hsu , Yiming Zhang This is my paper

Pith reviewed 2026-05-10 10:23 UTC · model grok-4.3

classification 💻 cs.DS cs.NI
keywords IPv6 lookuplongest prefix matchB+ treepacket forwardingIP lookupdata structuresvectorizationsoftware performance
0
0 comments X

The pith

IPv6 longest-prefix lookup can be reframed as one-dimensional search over elementary intervals using a linearized B+-tree to reach 390 million lookups per second on one core.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tries to establish that the longest prefix match problem in IPv6 forwarding, which mixes address values with varying prefix lengths, performs poorly when split into separate one-dimensional searches. By instead mapping the entire space to elementary intervals that capture all possible matches in a single dimension, the authors create a unified search space. They then build a linearized B+-tree on this space and equip it with vectorized, branch-free operations to exploit modern CPU features. A sympathetic reader would care because high-speed software lookup matters for cost-effective routers handling growing IPv6 tables with diverse prefix lengths, potentially avoiding expensive hardware accelerators. If successful, it shows that careful geometric transformation of the problem can yield substantial practical speedups.

Core claim

PlanB transforms the inherently two-dimensional longest prefix match into an equivalent one-dimensional search over elementary intervals, unifying prefix values and lengths. It adapts a flat-array B-tree into a linearized B+-tree tailored for this space and applies vectorization, batching, branch-free logic, and loop unrolling. On real-world IPv6 FIBs, this yields 390 million lookups per second on a single AMD core and scales to 3.4 billion on twelve cores, exceeding prior software schemes by factors of 1.6 to 14.

What carries the argument

The linearized B+-tree, a flat-array adaptation of the B+-tree structure fitted to the one-dimensional elementary interval representation of longest-prefix data.

If this is right

  • Software IPv6 routers can sustain hundreds of millions of lookups per core on commodity AMD processors.
  • Throughput scales nearly linearly across all available cores up to 3.4 billion lookups per second.
  • Memory usage remains practical while correctness holds for real-world forwarding information bases.
  • Prior schemes such as PopTrie, CP-Trie, Neurotrie, and HBS are outperformed by 1.6 to 14 times on the same data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The 2D-to-1D interval mapping technique could be applied to other range or prefix queries in databases and network security systems.
  • The same structure might be re-tested on mixed IPv4 and IPv6 tables or on ARM and Intel processors to check architectural portability.
  • Hardware offload of the vectorized search loop could push performance beyond current software limits.

Load-bearing premise

That the conversion of any set of IPv6 prefixes into elementary intervals produces a complete and correct one-dimensional ordering from which the longest matching prefix can always be recovered without omissions or conflicts.

What would settle it

An exhaustive test on synthetic IPv6 prefix collections that include every combination of overlapping lengths, verifying that every possible input address returns exactly the longest correct prefix and next-hop value.

Figures

Figures reproduced from arXiv: 2604.14650 by Chen Chen, Huiba Li, Jiwu Shu, Lanzheng Liu, Windsor Hsu, Yiming Zhang, Zhihao Zhang.

Figure 1
Figure 1. Figure 1: Lookup performance of four software schemes. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of IPv6 prefixes as prioritized address [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: An example of the flat array layout of a 3-ary B-tree. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: An example of a 3-ary linearized B +-tree. at offset j × b within level d + 1. Therefore, the starting index of the keys for a specific child c (where c is from 0 to b−1) of the j-th node at level d is given by: child_start(d, j, c) = level_start(d +1) + (j ×b+c)×k Parameter Tuning. We tune b and k for cache efficiency, sizing each node to fit a 64-byte cache line (common on x86 CPUs). PlanB uses 64-bit ke… view at source ↗
Figure 6
Figure 6. Figure 6: PlanB’s RX-LPM-TX module within DPDK. path. Second, the transition is seamless. Once the new struc￾ture is ready, a single atomic pointer swap redirects all new lookups. This guarantees that every lookup operation com￾pletes correctly without being affected by the update. Third, by batching updates and performing a full rebuild, PlanB amortizes the update cost and efficiently handles high-volume or bursty … view at source ↗
Figure 7
Figure 7. Figure 7: System throughput for DPDK-based L3 forwarding. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Lookup speed with single core on Intel CPU. [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Lookup speed with multi-cores on Intel CPU. [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Memory overhead. world performance is bottlenecked by high per-node overhead. Each step in its lookup requires chasing pointers between sep￾arate memory structures and executing complex scalar logic. HBS’s performance is sensitive to the number of distinct pre￾fix lengths and the efficiency of its hash function. Multi-Core Scalability. Figs. 10 and 11 show the aggre￾gated lookup throughput as the number o… view at source ↗
Figure 14
Figure 14. Figure 14: Ablation study of PlanB’s design components. [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
read the original abstract

IP lookup via Longest Prefix Match (LPM) is critical for packet forwarding. Unfortunately, conventional lookup algorithms are inefficient for IPv6 Forwarding Information Bases (FIBs), which are characterized by a set of long prefixes with diverse lengths. We observe that LPM inherently represents a two-dimensional (2D) search problem over both prefix values and prefix lengths, but existing algorithms mostly treat LPM as two separate levels of one-dimensional (1D) searches, causing poor lookup performance and high memory overhead. This paper presents PlanB, a novel scheme for high-speed IPv6 lookup. We transform the 2D LPM into an equivalent 1D search problem over elementary intervals, thereby unifying the search across prefix value and lengths. We then adapt a flat-array-based B-tree structure to the needs of LPM to propose the linearized $B^+$-tree, based on which we introduce an efficient search algorithm tailored to the properties of the transformed space. To maximize performance, we integrate PlanB with vectorization, batching, branch-free logic, and loop unrolling to fully exploit CPU parallelism. Extensive evaluation shows that PlanB achieves single-core performance of 390 Million Lookups Per Sec (MLPS) with real-world IPv6 FIBs on AMD processor, and scales to full-12-core performance of 3.4 Billion Lookups Per Sec (BLPS). This is 1.6$\times$$\sim$14$\times$ higher than state-of-the-art software-based schemes (PopTrie, CP-Trie, Neurotrie and HBS).

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript presents PlanB, a scheme for high-speed software IPv6 longest-prefix-match (LPM) lookup. It transforms the inherently 2D LPM problem (over prefix values and lengths) into an equivalent 1D search over elementary intervals, introduces a linearized B⁺-tree data structure with a tailored search algorithm, and applies vectorization, batching, branch-free logic, and loop unrolling to achieve 390 MLPS single-core and 3.4 BLPS on 12 cores on real-world IPv6 FIBs—1.6×–14× faster than PopTrie, CP-Trie, Neurotrie, and HBS.

Significance. If the 2D-to-1D equivalence holds and the reported throughputs are reproducible on production hardware, PlanB would represent a meaningful advance in software-based IPv6 forwarding, particularly for high-speed routers and NFV environments where memory efficiency and cache locality matter. The engineering focus on CPU-specific optimizations (vectorization, unrolling) is a practical strength that could influence future lookup designs.

major comments (3)
  1. [§3] §3 (2D-to-1D Transformation): The central claim that converting LPM into a 1D search over elementary intervals always yields the correct longest prefix relies on the intervals being disjoint, exhaustive, and ordered such that the first match is the longest; the manuscript supplies a construction algorithm but no explicit invariant, inductive argument, or proof that this property holds for arbitrary IPv6 prefix sets containing nested or crossing prefixes of lengths 1–128.
  2. [§5] §5 (Evaluation): The headline figures of 390 MLPS (single-core) and 3.4 BLPS (12-core) are presented without accompanying memory-footprint measurements, cache-miss counts, FIB-size statistics, or error bars across multiple runs, which weakens the ability to verify that the linearized B⁺-tree plus vectorized search actually sustains the claimed performance without excessive memory or locality penalties.
  3. [§4] §4 (Linearized B⁺-tree Search): The description of the vectorized, branch-free search over the linearized structure does not explicitly address how the algorithm handles the case where multiple intervals could match (due to prefix nesting) while still guaranteeing the longest match, nor does it quantify the worst-case number of intervals generated for pathological FIBs.
minor comments (2)
  1. [Abstract] Abstract: LaTeX artifacts remain in the text (e.g., “linearized $B^+$-tree” and “1.6$×$∼14$×$”) that should be rendered cleanly for the final version.
  2. [§2] §2 (Related Work): The comparison table or discussion with PopTrie, CP-Trie, Neurotrie, and HBS would benefit from explicit column headings for memory usage and lookup latency to make the 1.6×–14× claim easier to interpret.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below, indicating where we will revise the manuscript and providing additional clarification where the existing text already supports the claims.

read point-by-point responses

  1. Referee: [§3] §3 (2D-to-1D Transformation): The central claim that converting LPM into a 1D search over elementary intervals always yields the correct longest prefix relies on the intervals being disjoint, exhaustive, and ordered such that the first match is the longest; the manuscript supplies a construction algorithm but no explicit invariant, inductive argument, or proof that this property holds for arbitrary IPv6 prefix sets containing nested or crossing prefixes of lengths 1–128.

    Authors: We agree that an explicit invariant and proof sketch would improve rigor. Section 3 constructs elementary intervals by sorting all prefix boundaries and assigning to each resulting interval the longest prefix that covers it. This guarantees disjointness and exhaustiveness by definition of the boundaries, and the ordering ensures that the first (rightmost) interval whose range contains the lookup key holds the longest match. We will add a formal invariant statement plus a short inductive argument over the sorted prefix list in the revised Section 3. revision: yes

  2. Referee: [§5] §5 (Evaluation): The headline figures of 390 MLPS (single-core) and 3.4 BLPS (12-core) are presented without accompanying memory-footprint measurements, cache-miss counts, FIB-size statistics, or error bars across multiple runs, which weakens the ability to verify that the linearized B⁺-tree plus vectorized search actually sustains the claimed performance without excessive memory or locality penalties.

    Authors: We concur that these supporting metrics strengthen reproducibility. In the revision we will augment Section 5 with memory footprints for each FIB, L1/L2/L3 cache-miss rates obtained via perf, exact FIB sizes and prefix-length histograms for the evaluated traces, and error bars (standard deviation) computed over ten independent runs. revision: yes

  3. Referee: [§4] §4 (Linearized B⁺-tree Search): The description of the vectorized, branch-free search over the linearized structure does not explicitly address how the algorithm handles the case where multiple intervals could match (due to prefix nesting) while still guaranteeing the longest match, nor does it quantify the worst-case number of intervals generated for pathological FIBs.

    Authors: Because the elementary intervals are stored in sorted order and are disjoint, the branch-free search simply locates the rightmost interval whose start value ≤ key; the construction invariant ensures this interval carries the longest original prefix. No additional matching step is required. The number of intervals is at most 2n−1 for n prefixes. We will state this handling explicitly and add the linear bound (together with observed expansion factors on real IPv6 FIBs) in the revised Section 4. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces a new algorithmic construction that transforms 2D LPM into an equivalent 1D search over elementary intervals and then applies a linearized B+-tree with vectorized optimizations. No equations, fitted parameters, or self-citations are shown to reduce the central claims to their own inputs by construction. The performance numbers are presented as empirical results from evaluation on real-world IPv6 FIBs rather than predictions forced by prior fits or definitions. The derivation is therefore self-contained and externally falsifiable via implementation and benchmarking.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract-only review; the scheme rests on standard B+-tree search properties and the correctness of the 2D-to-1D interval transformation, neither of which is derived in the provided text.

axioms (1)
  • standard math B+-tree maintains sorted order and balanced height for logarithmic search
    Invoked implicitly when the linearized structure is used for fast lookup.
invented entities (1)
  • linearized B+-tree no independent evidence
    purpose: Flat-array B+-tree variant adapted to the elementary-interval representation of LPM
    New data structure introduced to unify prefix value and length search.

pith-pipeline@v0.9.0 · 5605 in / 1318 out tokens · 32065 ms · 2026-05-10T10:23:43.655925+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages

  1. [1]

    ht tps://ris.ripe.net/docs/route-collectors/

    RIPE RIS is a BGP routing data collection platform. ht tps://ris.ripe.net/docs/route-collectors/

    work page

  2. [2]

    https:// www.routeviews.org/routeviews/collectors/

    University of Oregon RouteViews Project. https:// www.routeviews.org/routeviews/collectors/

    work page

  3. [3]

    Poptrie: A com- pressed trie with population count for fast and scalable software ip routing table lookup

    Hirochika Asai and Yasuhiro Ohara. Poptrie: A com- pressed trie with population count for fast and scalable software ip routing table lookup. InProceedings of the 2015 ACM Conference on Special Interest Group on Data Communication (SIGCOMM’15), pages 57–70, 2015

    work page 2015

  4. [4]

    Jonathan Chao

    Masanori Bando, Yi-Li Lin, and H. Jonathan Chao. Flashtrie: Beyond 100-gb/s ip route lookup using hash- based prefix-compressed trie.IEEE/ACM Transactions on Networking, 20(4):1262–1275, 2012

    work page 2012

  5. [5]

    Addressing tcam limi- tations of software-defined networks for content-based routing

    Sukanya Bhowmik, Muhammad Adnan Tariq, Alexan- der Balogh, and Kurt Rothermel. Addressing tcam limi- tations of software-defined networks for content-based routing. InProceedings of the 11th ACM International Conference on Distributed and Event-Based Systems (DEBS’17), pages 100–111, 2017

    work page 2017

  6. [6]

    https://stats.labs .apnic.net/ipv6

    IPv6 Capable Rate by Country. https://stats.labs .apnic.net/ipv6

    work page

  7. [7]

    Rethinking packet forwarding hardware

    Martin Casado, Teemu Koponen, Daekyeong Moon, and Scott Shenker. Rethinking packet forwarding hardware. InHotNets, pages 1–6, 2008

    work page 2008

  8. [8]

    Scaling ip lookup to large databases using the cram lens

    Robert Chang, Pradeep Dogga, Andy Fingerhut, Vic- tor Rios, and George Varghese. Scaling ip lookup to large databases using the cram lens. InProceedings of the 22nd USENIX Symposium on Networked Sys- tems Design and Implementation, NSDI ’25, USA, 2025. USENIX Association

    work page 2025

  9. [9]

    Cat- cam: Constant-time alteration ternary cam with scal- able in-memory architecture

    Dibei Chen, Zhaoshi Li, Tianzhu Xiong, Zhiwei Liu, Jun Yang, Shouyi Yin, Shaojun Wei, and Leibo Liu. Cat- cam: Constant-time alteration ternary cam with scal- able in-memory architecture. In2020 53rd Annual IEEE/ACM International Symposium on Microarchitec- ture (MICRO’20), pages 342–355, 2020

    work page 2020

  10. [10]

    utree: a persistent b+-tree with low tail latency.Proc

    Youmin Chen, Youyou Lu, Kedong Fang, Qing Wang, and Jiwu Shu. utree: a persistent b+-tree with low tail latency.Proc. VLDB Endow., pages 2634–2648, July 2020

    work page 2020

  11. [11]

    Lu- ori: Active probing and evaluation of internet-wide ipv6 fully responsive prefixes

    Daguo Cheng, Lin He, Chentian Wei, Qilei Yin, Boran Jin, Zhaoan Wang, Xiaoteng Pan, Sixu Zhou, Ying Liu, Shenglin Zhang, Fuchao Tan, and Wenmao Liu. Lu- ori: Active probing and evaluation of internet-wide ipv6 fully responsive prefixes. In2024 IEEE 32nd Inter- national Conference on Network Protocols (ICNP’24), pages 1–12, 2024

    work page 2024

  12. [12]

    The 55th statistical report on china’s internet develop- ment

    China Internet Network Information Center (CNNIC). The 55th statistical report on china’s internet develop- ment. https://www.cnnic.com.cn/IDR/ReportDow nloads/202505/P020250514564119130448.pdf

    work page

  13. [13]

    Ubiquitous b-tree.ACM Comput

    Douglas Comer. Ubiquitous b-tree.ACM Comput. Surv., 11(2):121–137, June 1979

    work page 1979

  14. [14]

    https://bgp.potaroo.net/ v6/as2.0/index.html

    IPv6 BGP Table Data. https://bgp.potaroo.net/ v6/as2.0/index.html

    work page

  15. [15]

    Routebricks: exploiting parallelism to scale software routers

    Mihai Dobrescu, Norbert Egi, Katerina Argyraki, Byung-Gon Chun, Kevin Fall, Gianluca Iannaccone, Allan Knies, Maziar Manesh, and Sylvia Ratnasamy. Routebricks: exploiting parallelism to scale software routers. InProceedings of the ACM SIGOPS 22nd Sym- posium on Operating Systems Principles (SOSP’09), pages 15–28, 2009

    work page 2009

  16. [16]

    https://www.inte l.com/content/www/us/en/content-details/6 71368/intel-architecture-instruction-set-e xtensions-programming-reference.html

    Intel Architecture Instruction Set Extensions and Future Features Programming Reference. https://www.inte l.com/content/www/us/en/content-details/6 71368/intel-architecture-instruction-set-e xtensions-programming-reference.html

    work page

  17. [17]

    Metronome: Adaptive and precise intermittent packet retrieval in dpdk.IEEE/ACM Transactions on Networking, 31(3):979–993, 2023

    Marco Faltelli, Giacomo Belocchi, Francesco Quaglia, Salvatore Pontarelli, and Giuseppe Bianchi. Metronome: Adaptive and precise intermittent packet retrieval in dpdk.IEEE/ACM Transactions on Networking, 31(3):979–993, 2023

    work page 2023

  18. [18]

    Andrew D. Ferguson, Steve Gribble, Chi-Yao Hong, Charles Killian, Waqar Mohsin, Henrik Muehe, Joon Ong, Leon Poutievski, Arjun Singh, Lorenzo Vicisano, Richard Alimi, Shawn Shuoshuo Chen, Mike Conley, Subhasree Mandal, Karthik Nagaraj, Kondapa Naidu Bollineni, Amr Sabaa, Shidong Zhang, Min Zhu, and Amin Vahdat. Orion: Google’s Software-Defined net- workin...

    work page 2021

  19. [19]

    https://de veloper.apple.com/documentation/apple-sil icon/cpu-optimization-guide

    Apple Silicon CPU Optimization Guide. https://de veloper.apple.com/documentation/apple-sil icon/cpu-optimization-guide

    work page

  20. [20]

    Achieving high utilization with software-driven wan

    Chi-Yao Hong, Srikanth Kandula, Ratul Mahajan, Ming Zhang, Vijay Gill, Mohan Nanduri, and Roger Watten- hofer. Achieving high utilization with software-driven wan. InProceedings of the ACM SIGCOMM 2013 Con- ference on SIGCOMM, pages 15–26, 2013

    work page 2013

  21. [21]

    https://www.cac.gov

    IPv6 Development in China. https://www.cac.gov. cn/2025-08/1/c_1755590302116970.htm

    work page 2025

  22. [22]

    C2rtl: A high- level synthesis system for ip lookup and packet classifi- cation

    Md Iftakharul Islam and Javed I Khan. C2rtl: A high- level synthesis system for ip lookup and packet classifi- cation. In2021 IEEE 22nd International Conference on High Performance Switching and Routing (HPSR’21), pages 1–8, 2021

    work page 2021

  23. [23]

    Cp-trie: Cumu- lative popcount based trie for ipv6 routing table lookup in software and asic

    Md Iftakharul Islam and Javed I Khan. Cp-trie: Cumu- lative popcount based trie for ipv6 routing table lookup in software and asic. In2021 IEEE 22nd International Conference on High Performance Switching and Rout- ing (HPSR’21), pages 1–8, 2021

    work page 2021

  24. [24]

    Heuristic binary search: Adaptive and fast ipv6 route lookup with incremental up- dates

    Donghong Jiang, Yanbiao Li, Yuxuan Chen, Jing Hu, Yi Huang, and Gaogang Xie. Heuristic binary search: Adaptive and fast ipv6 route lookup with incremental up- dates. InProceedings of the 7th Asia-Pacific Workshop on Networking (APNet’23), pages 47–53, 2023

    work page 2023

  25. [25]

    Heuristic binary search: Adaptive and fast ipv6 route lookup with incremental prefix updates.IEEE Transactions on Networking, pages 554–569, 2025

    Donghong Jiang, Yanbiao Li, Yuxuan Chen, Jing Hu, Yi Huang, and Gaogang Xie. Heuristic binary search: Adaptive and fast ipv6 route lookup with incremental prefix updates.IEEE Transactions on Networking, pages 554–569, 2025

    work page 2025

  26. [26]

    Jiang, Q

    W. Jiang, Q. Wang, and V . K. Prasanna. Beyond tcams: An sram-based parallel multi-pipeline architecture for terabit ip lookup. InProceedings of the IEEE INFO- COM 2008, pages 1786–1794, 2008

    work page 2008

  27. [27]

    Stateless network functions

    Murad Kablan, Blake Caldwell, Richard Han, Hani Jamjoom, and Eric Keller. Stateless network functions. InProceedings of the 2015 ACM SIGCOMM Workshop on Hot Topics in Middleboxes and Network Function Virtualization, pages 49–54, 2015

    work page 2015

  28. [28]

    Array layouts for comparison-based searching.ACM J

    Paul-Virak Khuong and Pat Morin. Array layouts for comparison-based searching.ACM J. Exp. Algorithmics, 22, May 2017

    work page 2017

  29. [29]

    The adaptive radix tree: Artful indexing for main-memory databases

    Viktor Leis, Alfons Kemper, and Thomas Neumann. The adaptive radix tree: Artful indexing for main-memory databases. InProceedings of the 2013 IEEE Inter- national Conference on Data Engineering (ICDE’13), pages 38–49, 2013

    work page 2013

  30. [30]

    Dhl: Enabling flexible software network functions with fpga acceleration

    Xiaoyao Li, Xiuxiu Wang, Fangming Liu, and Hong Xu. Dhl: Enabling flexible software network functions with fpga acceleration. In2018 IEEE 38th International Con- ference on Distributed Computing Systems (ICDCS’18), pages 1–11, 2018

    work page 2018

  31. [31]

    An ipv6 routing lookup algorithm based on hash table and hot

    Yuqiang Li, Fan Zhou, Xiaoxiang Zhu, and Jianming Liao. An ipv6 routing lookup algorithm based on hash table and hot. In2022 5th International Conference on Information Communication and Signal Processing (ICICSP), pages 397–402, 2022

    work page 2022

  32. [32]

    https://doc.dp dk.org/guides-16.04/sample_app_ug/perform ance_thread.html

    DPDK Performance Thread Model. https://doc.dp dk.org/guides-16.04/sample_app_ug/perform ance_thread.html

    work page

  33. [33]

    Mogul and John Wilkes

    Jeffrey C. Mogul and John Wilkes. Physical deployabil- ity matters. InProceedings of the 22nd ACM Workshop on Hot Topics in Networks (HotNets’23), pages 9–17, 2023

    work page 2023

  34. [34]

    Md S. Q. Zulkar Nine, Tevfik Kosar, Muhammed Fatih Bulut, and Jinho Hwang. Greennfv: Energy-efficient network function virtualization with service level agree- ment constraints. InProceedings of the International Conference for High Performance Computing, Network- ing, Storage and Analysis (SC’23), 2023

    work page 2023

  35. [35]

    The multiple roles that ipv6 addresses can play in today’s internet.SIG- COMM Comput

    Maxime Piraux, Tom Barbette, Nicolas Rybowski, Louis Navarre, Thomas Alfroy, Cristel Pelsser, François Michel, and Olivier Bonaventure. The multiple roles that ipv6 addresses can play in today’s internet.SIG- COMM Comput. Commun. Rev., pages 10–18, 2022

    work page 2022

  36. [36]

    Neurolpm - scaling longest prefix match hardware with neural networks

    Alon Rashelbach, Igor de Paula, and Mark Silberstein. Neurolpm - scaling longest prefix match hardware with neural networks. InProceedings of the 56th Annual IEEE/ACM International Symposium on Microarchitec- ture (MICRO’23), pages 886–899, 2023

    work page 2023

  37. [37]

    https://www.in tel.com/content/dam/develop/external/us/en /documents/d9156103-705230.pdf

    Intel SSE4 Programming Reference. https://www.in tel.com/content/dam/develop/external/us/en /documents/d9156103-705230.pdf

    work page

  38. [38]

    Bento: safely bringing network function virtualization to tor

    Michael Reininger, Arushi Arora, Stephen Herwig, Nicholas Francino, Jayson Hurst, Christina Garman, and Dave Levin. Bento: safely bringing network function virtualization to tor. InProceedings of the 2021 ACM SIGCOMM 2021 Conference, pages 821–835, 2021

    work page 2021

  39. [39]

    Ipv6 hitlists at scale: Be careful what you wish for

    Erik Rye and Dave Levin. Ipv6 hitlists at scale: Be careful what you wish for. InProceedings of the ACM SIGCOMM 2023 Conference, pages 904–916, 2023

    work page 2023

  40. [40]

    How much tcam do we need for splitting traffic? InPro- ceedings of the ACM SIGCOMM Symposium on SDN Research (SOSR), pages 169–175, 2021

    Yaniv Sadeh, Ori Rottenstreich, and Haim Kaplan. How much tcam do we need for splitting traffic? InPro- ceedings of the ACM SIGCOMM Symposium on SDN Research (SOSR), pages 169–175, 2021

    work page 2021

  41. [41]

    Accessing cloud with disaggregated Software-Defined router

    Hua Shao, Xiaoliang Wang, Yuanwei Lu, Yanbo Yu, Shengli Zheng, and Youjian Zhao. Accessing cloud with disaggregated Software-Defined router. In18th USENIX Symposium on Networked Systems Design and Implementation (NSDI’21), pages 1–14, 2021

    work page 2021

  42. [42]

    High-performance ipv6 lookup with real-time updates using hierarchical-balanced search tree

    Tong Shen, Xian Yu, Gaogang Xie, and Dafang Zhang. High-performance ipv6 lookup with real-time updates using hierarchical-balanced search tree. In2018 IEEE Global Communications Conference (GLOBECOM’18), pages 1–7, 2018

    work page 2018

  43. [43]

    AddrMiner: A comprehensive global active IPv6 address discovery system

    Guanglei Song, Jiahai Yang, Lin He, Zhiliang Wang, Guo Li, Chenxin Duan, Yaozhong Liu, and Zhongxiang Sun. AddrMiner: A comprehensive global active IPv6 address discovery system. In2022 USENIX Annual Technical Conference (ATC’22), pages 309–326, 2022

    work page 2022

  44. [44]

    Ptcam: Scal- able high-speed name prefix lookup using tcam

    Tian Song, Tianlong Li, and Yating Yang. Ptcam: Scal- able high-speed name prefix lookup using tcam. In Proceedings of the ACM SIGCOMM 2025 Conference, SIGCOMM ’25, pages 707–719, 2025

    work page 2025

  45. [45]

    https://www.google.com/i ntl/en/ipv6/statistics.html

    Google IPv6 Statistics. https://www.google.com/i ntl/en/ipv6/statistics.html

    work page

  46. [46]

    Ship: A scalable high- performance ipv6 lookup algorithm that exploits prefix characteristics.IEEE/ACM Transactions on Networking, 27(4):1529–1542, 2019

    Thibaut Stimpfling, Normand Belanger, JM Pierre Lan- glois, and Yvon Savaria. Ship: A scalable high- performance ipv6 lookup algorithm that exploits prefix characteristics.IEEE/ACM Transactions on Networking, 27(4):1529–1542, 2019

    work page 2019

  47. [47]

    Pierre Langlois, Normand Bélanger, and Yvon Savaria

    Thibaut Stimpfling, J.M. Pierre Langlois, Normand Bélanger, and Yvon Savaria. A low-latency memory- efficient ipv6 lookup engine implemented on fpga using high-level synthesis. In2018 18th IEEE/ACM Interna- tional Symposium on Cluster, Cloud and Grid Comput- ing (CCGRID’18), pages 402–411, 2018

    work page 2018

  48. [48]

    Nfp: Enabling network function paral- lelism in nfv

    Chen Sun, Jun Bi, Zhilong Zheng, Heng Yu, and Hongxin Hu. Nfp: Enabling network function paral- lelism in nfv. InProceedings of the Conference of the ACM Special Interest Group on Data Communication (SIGCOMM’17), pages 43–56, 2017

    work page 2017

  49. [49]

    Varghese, and P.R

    Subhash Suri, G. Varghese, and P.R. Warkhede. Mul- tiway range trees: scalable ip lookup with fast updates. InGLOBECOM’01. IEEE Global Telecommunications Conference (Cat. No.01CH37270), volume 3, pages 1610–1614, 2001

    work page 2001

  50. [50]

    https://www.arm.com/te chnologies/neon

    ARM NEON Technology. https://www.arm.com/te chnologies/neon

    work page

  51. [51]

    ResQ: Enabling SLOs in network function vir- tualization

    Amin Tootoonchian, Aurojit Panda, Chang Lan, Melvin Walls, Katerina Argyraki, Sylvia Ratnasamy, and Scott Shenker. ResQ: Enabling SLOs in network function vir- tualization. In15th USENIX Symposium on Networked Systems Design and Implementation (NSDI’18), pages 283–297, 2018

    work page 2018

  52. [52]

    T-cache: Efficient policy-based forwarding using small tcam

    Ying Wan, Haoyu Song, Yang Xu, Yilun Wang, Tian Pan, Chuwen Zhang, Yi Wang, and Bin Liu. T-cache: Efficient policy-based forwarding using small tcam. IEEE/ACM Transactions on Networking, pages 2693– 2708, 2021

    work page 2021

  53. [53]

    Deft: A Scalable Tree Index for Disaggregated Mem- ory

    Jing Wang, Qing Wang, Yuhao Zhang, and Jiwu Shu. Deft: A Scalable Tree Index for Disaggregated Mem- ory. InProceedings of the Twentieth European Con- ference on Computer Systems (EuroSys’25), pages 886– 901, 2025

    work page 2025

  54. [54]

    Sherman: A write-optimized distributed b+tree index on disaggre- gated memory

    Qing Wang, Youyou Lu, and Jiwu Shu. Sherman: A write-optimized distributed b+tree index on disaggre- gated memory. InProceedings of the 2022 International Conference on Management of Data (SIGMOD’22), pages 1033–1048, 2022

    work page 2022

  55. [55]

    Multiway range trees: scalable ip lookup with fast updates.Comput

    Priyank Warkhede, Subhash Suri, and George Vargh- ese. Multiway range trees: scalable ip lookup with fast updates.Comput. Netw., 44(3):289–303, February 2004

    work page 2004

  56. [56]

    Liu, Qi Li, and Laurent Mathy

    Tong Yang, Gaogang Xie, YanBiao Li, Qiaobin Fu, Alex X. Liu, Qi Li, and Laurent Mathy. Guarantee ip lookup performance with fib explosion. InProceedings of ACM Conference on SIGCOMM (SIGCOMM’14), pages 39–50, 2014

    work page 2014

  57. [57]

    Liu, Qiaobin Fu, Yanbiao Li, Xiaoming Li, and Laurent Mathy

    Tong Yang, Gaogang Xie, Alex X. Liu, Qiaobin Fu, Yanbiao Li, Xiaoming Li, and Laurent Mathy. Constant ip lookup with fib explosion.IEEE/ACM Transactions on Networking, pages 1821–1836, 2018

    work page 2018

  58. [58]

    Tar: Traffic adaptive ipv6 routing lookup scheme

    Xinyi Zhang, Zhiyuan Xu, Huaiyi Zhao, Yanbiao Li, and Gaogang Xie. Tar: Traffic adaptive ipv6 routing lookup scheme. InProceedings of the 8th Asia-Pacific Workshop on Networking (APNet’24), pages 135–141, 2024

    work page 2024

  59. [59]

    A tcam-based distributed parallel ip lookup scheme and performance analysis.IEEE/ACM Transactions on Net- working, 14(4):863–875, 2006

    Kai Zheng, Chengchen Hu, Hongbin Lu, and Bin Liu. A tcam-based distributed parallel ip lookup scheme and performance analysis.IEEE/ACM Transactions on Net- working, 14(4):863–875, 2006

    work page 2006

  60. [60]

    Fast software ipv6 lookup with neurotrie.IEEE/ACM Transactions on Networking, pages 4040–4055, 2024

    Yuxi Zhu, Hao Chen, Yuan Yang, Mingwei Xu, Yuxuan Zhang, Chenyi Liu, and Jianping Wu. Fast software ipv6 lookup with neurotrie.IEEE/ACM Transactions on Networking, pages 4040–4055, 2024

    work page 2024