A 2016 simulation study (Argibay-Losada, Yoshida, Maruta, and Kitayama, J. Opt. Commun. Netw. 8(11):B43-B54) argued the counterintuitive result that metropolitan 5G networks built from bufferless optical packet switches (OPS) can outperform state-of-the-art electronic packet switching (EPS) for applications requiring reliable, low-latency delivery: lower flow completion times, fewer missed hard per-packet deadlines, and more packets outracing their EPS counterparts, with the advantage growing with network load. The claimed mechanism is that the absence of queueing delays gives end-to-end packet delays low mean and variance, letting transport-layer congestion control react swiftly to contention losses. The result, however, rests on idealizations: negligible switch processing/transmission times, no access-network dynamics, bidirectional-ring topologies, a narrow parameter sweep, and a single loss-tolerant congestion-control design (TCP stop-and-wait). The open problem is to map the safe operational envelope of bufferless OPS against EPS for delay-sensitive 5G traffic under realistic conditions - non-negligible per-switch latencies and access bottlenecks, a broader congestion-control design space, topologies and traffic beyond bidirectional rings, the heavy tail of retransmission-induced delay and its mitigation, and quantified cost/energy/reliability trade-offs. No prior work is known to map this safe operational envelope.

Background

In packet-switched networks, routers and switches forward self-contained datagrams and resolve contention (two packets needing the same output link at the same time) statistically rather than by pre-scheduled reservation as in circuit switching. Current electronic packet switching (EPS) converts incoming optical signals to the electronic domain inside each switch and stores contending packets in random-access-memory (RAM) buffers, so losses are rare (typically under 1%) but queueing delays and their variance accumulate at every hop. Optical packet switching (OPS) keeps packets in optical form inside the switch fabric, avoiding costly optical-electronic-optical conversion; because practical optical RAM does not exist, however, an OPS switch cannot store a contending packet. Proposed remedies include fiber delay lines (FDLs), fiber loops of different lengths that delay packets to emulate buffering, and hybrid OPS/EPS architectures, but both add significant complexity. A bufferless OPS switch simply discards the contending packet, so datagram loss rates are high even without congestion.

The source study considered metropolitan-area networks as a demanding 5G use case (5G denoting fifth-generation mobile networks, whose delay-sensitive services these metropolitan networks would carry: interconnecting cars and traffic-control elements, urban lighting and energy control, mobile traffic management), where many applications want a flow - a group of bytes - delivered reliably and as fast as possible. Its performance metrics were: flow completion time (time until the last byte is delivered losslessly, which retransmission by the Transmission Control Protocol (TCP), the Internet's standard reliable-transport protocol, makes possible), the probability of missing a hard per-packet end-to-end deadline (a real-time requirement), and the number of packets that outrun their counterparts in the competing network (winner-takes-all applications, such as latency-sensitive trading). By simulation on bidirectional ring metropolitan topologies, treating switch-internal switching and processing times as negligible and using a stop-and-wait transport protocol (TCP SAW, a single-outstanding-segment automatic-repeat variant the same first author had shown to boost throughput over fast optical switching at INFOCOM 2014), the study found bufferless OPS beats EPS on all three metrics, with the gap widening as load grows: contention losses are detected and repaired quickly because every packet experiences only propagation delay, whereas EPS losses from buffer overflow arrive late and after queueing has already inflated delays. A companion study the same year (J. Opt. Commun. Netw. 8(12):947) further developed transport-layer control to raise throughput in bufferless OPS networks.

The result is deliberately idealized. Switch-internal processing/transmission latencies and access-network segments between end users and the metropolitan core were assumed negligible; the conclusions were shown only for TCP SAW among the much larger space of congestion-control algorithms (variants for mixed round-trip times, multipath, heterogeneous loss patterns); the analysis covered bidirectional rings and selected parameter ranges (port counts, link rates, maximum transmission units, file-size mixes) rather than fat-tree local-area topologies (a switch topology built from many parallel, equal-cost paths, common in data centers) or wide-area mesh topologies, large switch fabrics, heterogeneous link rates, and real traffic traces; and the heavier tail of retransmission-induced delays that OPS exhibited empirically - a concern for services sensitive to worst-case (tail) latency - was observed but neither characterized nor mitigated (candidate mitigations include selective buffering at ingress/egress and adaptive retransmission timers). Cost, energy, reliability, and hardware-maturity factors for large-scale OPS fabrics were left unquantified. Whether the OPS advantage survives these realistic conditions - that is, the safe operational envelope for bufferless OPS in delay-sensitive 5G networks - is the open question. It matters against the field's broader history: OPS research peaked around 2004-2009 and then declined, with retrospective analyses attributing the failure to economics and hardware maturity rather than to any settled negative verdict on the delay/throughput performance question.

Problem Statement

Determine the safe operational envelope of bufferless optical packet switching (OPS) relative to electronic packet switching (EPS) for delay-sensitive 5G metropolitan networks - that is, establish, by analysis, simulation, and/or prototype measurement, under which realistic conditions the claimed OPS advantages (lower flow completion times, lower probability of missing hard per-packet end-to-end deadlines, more packets outrunning their EPS counterparts, higher throughput under reliable delivery) persist, and under which they disappear. Mapping this envelope spans the following open aspects, corresponding to the idealizations of the source study described above: (1) the effect of non-negligible per-switch processing and transmission latencies in OPS and EPS switches and of realistic access-network bottlenecks between end users and the metropolitan core on the OPS-versus-EPS trade-offs; (2) the space of end-to-end congestion-control designs beyond the stop-and-wait example used in the source (including variants for mixed round-trip times, multipath, and heterogeneous loss patterns), their tuning for bufferless OPS, and whether the OPS advantage is robust to this choice; (3) generalization beyond bidirectional rings - fat-tree local-area topologies, wide-area meshes, larger switch port counts, heterogeneous link rates, and real-world traffic traces - at loads and file-size mixes beyond the source's selected parameter ranges; (4) characterization of the heavy tail of retransmission-induced delays observed under OPS (e.g., its quantiles as a function of load and topology), its operational impact on tail-latency-sensitive services, and possible mitigation strategies such as selective buffering at ingress/egress and adaptive retransmission timers; and (5) the deployment-facing factors - cost, energy, reliability, and the engineering feasibility of large-scale OPS switch fabrics - whose quantification would allow OPS and EPS deployments to be compared beyond the qualitative retrospective verdicts the field has since reached.

References

  1. Pablo Jesus Argibay-Losada, Yuki Yoshida, Akihiro Maruta, Ken-ichi Kitayama, 'Optical Versus Electronic Packet Switching in Delay-Sensitive 5G Networks: Myths Versus Advantages', Journal of Optical Communications and Networking 8(11):B43-B54, 2016. DOI: 10.1364/jocn.8.000b43. https://doi.org/10.1364/jocn.8.000b43
  2. Pablo Jesus Argibay-Losada, Yuki Yoshida, Akihiro Maruta, Ken-ichi Kitayama, 'Optical vs. electronic packet switching: myth and advantage in 5G time-critical applications', 2016 International Conference on Optical Network Design and Modeling (ONDM), pp. 1-6, 2016 (companion conference version of the source paper). DOI: 10.1109/ondm.2016.7494067. https://doi.org/10.1109/ondm.2016.7494067
  3. Pablo Jesus Argibay-Losada, Kseniia Nozhnina, Gokhan Sahin, Chunming Qiao, 'Using stop-and-wait to improve TCP throughput in fast optical switching (FOS) networks over short physical distances', IEEE INFOCOM 2014 - IEEE Conference on Computer Communications, pp. 1312-1320, 2014. DOI: 10.1109/infocom.2014.6848064. https://doi.org/10.1109/infocom.2014.6848064
  4. Pablo Jesus Argibay-Losada, Gokhan Sahin, Kseniia Nozhnina, Chunming Qiao, 'Transport-Layer Control to Increase Throughput in Bufferless Optical Packet-Switching Networks', Journal of Optical Communications and Networking 8(12):947, 2016. DOI: 10.1364/jocn.8.000947. https://doi.org/10.1364/jocn.8.000947
  5. Pablo Jesus Argibay-Losada, Dominique Chiaroni, Chunming Qiao, 'Optical Packet Switching and Optical Burst Switching', Springer Handbook of Optical Networks, pp. 665-701, 2020. DOI: 10.1007/978-3-030-16250-4_20. https://doi.org/10.1007/978-3-030-16250-4_20
  6. Jose Roberto de Almeida Amazonas, German Santos-Boada, Josep Sole-Pareta, 'Who shot Optical Packet switching?', 2017 19th International Conference on Transparent Optical Networks (ICTON), pp. 1-4, 2017. DOI: 10.1109/icton.2017.8025164. https://doi.org/10.1109/icton.2017.8025164
  7. Franco Callegati, Davide Careglio, Luiz Henrique Bonani, Mario Pickavet, Josep Sole-Pareta, 'Why Optical Packet Switching failed and can Elastic Optical Networks take its place?', Optical Switching and Networking 44:100664, 2022. DOI: 10.1016/j.osn.2021.100664. https://doi.org/10.1016/j.osn.2021.100664
  8. Hideki Tode, Yuusuke Hashimoto, Yosuke Tanigawa, Yusuke Hirota, 'A Novel Transport Protocol Suitable for Bufferless Optical Packet Switching Networks', 2023 International Conference on Photonics in Switching and Computing (PSC), pp. 1-3, 2023. DOI: 10.1109/psc57974.2023.10297151. https://doi.org/10.1109/psc57974.2023.10297151
  9. Keishi Nishio, Yosuke Tanigawa, Yusuke Hirota, Hideki Tode, 'Quantitative Performance Evaluation of Microburst Packet Transmission Control for Transport Layer Protocols in Optical Packet Switching Networks', 2025 IEEE 26th International Conference on High Performance Switching and Routing (HPSR), pp. 1-5, 2025. DOI: 10.1109/hpsr64165.2025.11038867. https://doi.org/10.1109/hpsr64165.2025.11038867
  10. Mirosław Kantor, Edyta Biernacka, Piotr Boryło, Jerzy Domżał, Piotr Jurkiewicz, Miłosz Stypiński, Robert Wójcik, 'A survey on multi-layer IP and optical Software-Defined Networks', Computer Networks 162:106844, 2019. DOI: 10.1016/j.comnet.2019.06.022. https://doi.org/10.1016/j.comnet.2019.06.022
  11. I-Fen Chao, Kai-Yuan Hsu, 'High-performance system design for a large-scale AWG-based WDM star network', Journal of Optical Communications and Networking 17(6):459, 2025. DOI: 10.1364/JOCN.557451. https://doi.org/10.1364/JOCN.557451
  12. I-Fen Chao, Chain-Hung Lee, 'AWG-based WDM ring networks: High-performance and low-cost system designs', Computer Networks 145:64-75, 2018. DOI: 10.1016/j.comnet.2018.08.012. https://doi.org/10.1016/j.comnet.2018.08.012
  13. Boyang Zhou, Isaac Howenstine, Siraphob Limprapaipong, Liang Cheng, 'A Survey on Network Calculus Tools for Network Infrastructure in Real-Time Systems', IEEE Access 8:223588-223605, 2020. DOI: 10.1109/ACCESS.2020.3043600. https://doi.org/10.1109/ACCESS.2020.3043600
  14. Huasong Zeng, Chunxi Zhang, Xiaosu Yi, Rui Xiong, 'Heuristic fragmentation-aware scheduling for a multicluster time-sensitive passive optical LAN', Optical Fiber Technology 66:102662, 2021. DOI: 10.1016/j.yofte.2021.102662. https://doi.org/10.1016/j.yofte.2021.102662