Pith. sign in

REVIEW 3 major objections 5 minor 50 references

Congestion Management in High-Performance Interconnection Networks Using Adaptive Routing Notifications

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Adaptive routing notifications plus a dedicated buffer channel can eliminate congestion impact in fat-tree interconnects.

desk verdict A plausible, well-described simulation study of ARN+AFI congestion management with an overclaimed 'removes' headline and a load-bearing assumption about real InfiniBand ARN semantics. read the letter →

arxiv 2502.00616 v1 pith:F5YQKEY4 submitted 2025-02-02 cs.NI

Jose Rocher-Gonzalez , Jesus Escudero-Sahuquillo , Pedro J. Garcia , Francisco J. Quiles This is my paper · ORCID
classification cs.NI
keywords congestionmanagementadaptiveroutingnotificationsfat-treenetworksHead-of-LineblockingvirtualchannelsInfiniBandspreading
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that adaptive routing in lossless fat-tree networks can make congestion worse by spreading blocked flows across many paths, and that existing techniques based only on local switch information cannot stop this spreading. It proposes ARN+AFI, which uses the adaptive routing notification (ARN) mechanism found in some InfiniBand switches to tell upstream switches where a congestion tree actually starts, then marks and reroutes only the packets headed into that root and isolates them in a dedicated virtual channel, the Adapted Flow Channel. The simulations, run on 432-node and 3456-node fat trees under uniform, incast, multi-incast, and MPI-trace traffic, show that ARN+AFI recovers network efficiency within milliseconds where other routing and queuing schemes stay degraded. The paper's conclusion is that, in the scenarios evaluated, congestion impact is eliminated, at the price of a delay proportional to the congestion tree size.

What carries the argument

The machinery is the ARN table plus the AFC. The ARN table entry stores destination, output port, VC, a unique ARNid, congestionRootInfo (the fat-tree stage), and a consumed flag; it is what lets a switch decide whether an incoming packet should be adapted. The AFC is the reserved last virtual channel per input buffer, selected by the SL-to-VC table when the packet's adapted bit is set, and it is where all re-routed packets travel deterministically. The congestion detector (VOQ occupancy versus the high threshold, downstream free credits versus the credit threshold, and the congestion root timer) is what decides whether a port is a congestion root or a branch. Together these pieces turn a local congestion signal into a global rerouting decision with bounded spread.

What would settle it

A test on real HDR InfiniBand hardware: generate a single incast to one destination and watch an upstream switch's routing table; if no ARN-carrying packet with destination and VC information arrives upstream and triggers an alternative port selection, the mechanism's core assumption is false. A simpler simulator experiment that turns ARN generation off while keeping AFC isolation on would show how much of the reported recovery actually comes from the notifications.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that congestion-root information, propagated upstream through ARNs and combined with exact isolation of the adapted flows, fixes the main failure mode of adaptive routing: congestion spreading. A switch detects a root when a VOQ exceeds the high congestion detection threshold and the downstream port still has free credits above the free credits threshold for the congestion root timer, so it knows the point is a root rather than a branch. It then emits an ARN carrying the destination, VC, port, and congestion stage; switches along the upstream path forward it until a switch in a lower stage can consume it and select an alternative upward port. Packets matching the consumed entry are marked adapted, sent through the alternative route, and stored in the AFC at every downstream buffer until they reach the destination. This keeps non-congesting flows in regular VCs, avoiding Head-of-Line blocking and buffer hogging, while the rerouted flows travel deterministically in the AFC so they do not spawn new congestion trees. The stated result is that in all evaluated scenarios the congestion impact is eliminated, with a delay that grows with the congestion tree.

Load-bearing premise

The proposal depends on commercial InfiniBand switches generating, forwarding, consuming, and expiring ARNs exactly as the pseudocode models, behavior supported only by a patent and vendor documentation rather than public measurements.

Editorial extensions

If this is right

  • If the simulation results carry over to hardware, ARN+AFI can restore near-full network efficiency within milliseconds during incast congestion, while adaptive routing alone leaves efficiency near zero.
  • Combining ARN+AFI with topology-aware static queuing schemes such as vFtree or Flow2SL shortens the reaction time compared with DBBM, especially in larger networks.
  • ARN+AFI prevents congestion trees from delaying MPI applications: in PTRANS and Inception-v3 traces, runtime overhead from an injected incast is close to zero, where other techniques add hundreds of milliseconds.
  • The hardware changes required are small: an adapted bit in the packet header and SL-to-VL table logic to direct marked packets into the AFC, so the technique is implementable in InfiniBand HDR networks.
  • The technique also handles multiple simultaneous congestion trees, isolating each tree's flows in the AFC and recovering faster than single-root scenarios because more destinations drain the trees.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper evaluates only incast-style congestion; an obvious extension the authors mention as future work is in-network congestion, where ARN generation at internal points might behave differently because roots are harder to distinguish from branches.
  • Because the benefit hinges on ARN delivery latency, real hardware rate-limiting or loss of ARN packets could turn the millisecond recovery into a longer transient; a hardware-level experiment would settle this.
  • The AFC isolation logic is topology-agnostic, so the same idea could be tried on dragonflies or other topologies that support inter-switch congestion notifications, where the consumption stage rule would need to be replaced by another criterion.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes ARN+AFI, a congestion-management strategy for fat-tree interconnection networks that combines Adaptive Routing Notifications (ARNs), available in recent InfiniBand switches, with Adapted-Flow Isolation (AFI). The idea is to detect a congestion root by monitoring VOQ occupancy and link-credit thresholds, propagate that information upstream via ARN packets, and then reroute only the flows associated with the congested destination into a dedicated virtual channel (AFC), where they follow deterministic routing. The proposal is evaluated in the SAURON/OMNeT++ simulator on 432-node and 3456-node RLFTs using synthetic incast patterns (H10, H25, H10-4, H25-4) and two MPI traces (PTRANS and Inception-v3), combined with several static queuing schemes. The reported results show that ARN+AFI recovers network efficiency faster than the evaluated baselines and reduces the execution-time overhead of trace workloads under injected congestion. The central claim is that congestion impact is 'removed' or 'eliminated' in the evaluated scenarios.

Significance. If the results hold, ARN+AFI is a practical improvement for congestion management in fat-tree interconnects, building on ARN support already present in some commercial InfiniBand products. The paper provides useful implementation details (Pseudo-codes 1-4, ARN table fields, SL-to-VC mapping) and a comparatively broad evaluation across traffic patterns, queueing schemes, and two network sizes. The work also extends the authors' earlier AFI proposal by using inter-switch notifications to identify congesting flows more precisely. The significance is tempered by two factors: the entire mechanism depends on ARN semantics that are assumed rather than demonstrated on real switches, and the evaluation is simulation-only with the authors' own simulator and no independent hardware validation or parameter-sensitivity analysis. Nevertheless, the qualitative advantage of ARN+AFI over the baselines is visible in the plots and the central idea is defensible.

major comments (3)
  1. [Section 2.2 and Pseudo-codes 1-4] The practical validity of ARN+AFI rests on the assumption that commercial InfiniBand switches implement ARNs as multi-hop upstream notifications that can be forwarded through several switches, matched against per-destination routing-table entries, and consumed at a lower fat-tree stage. The paper cites the Mellanox patent [40] and vendor documentation [31,32] but provides no measurements, hardware experiments, or firmware-level evidence that these exact semantics exist in real devices. If actual ARNs are consumed only by the immediate upstream neighbor, or if they do not carry the destination/port/VC fields used in Pseudo-code 4, the simulated congestion-tree isolation cannot be delivered in practice. The implementation-detail discussion in Section 3.5 assumes the required behavior rather than demonstrating it.
  2. [Abstract and Section 5] The conclusion states that 'the congestion impact is eliminated in the evaluated scenarios,' and the abstract repeats that the proposal 'removes the congestion impact.' This is stronger than the evidence. The plots in Figures 6 and 7 show ARN+AFI experiencing a clear efficiency drop at hot-spot start—e.g., to approximately 0.1 in Figures 6e, 7e, and 7m—before recovering within a few milliseconds. The data support the claim that ARN+AFI mitigates congestion impact and recovers faster than the baselines, not that the impact is eliminated. Please soften the headline claim accordingly.
  3. [Section 4.1, congestion detector parameters] The four key thresholds HCDTh, LCDTh, FCTh, and CRT are fixed values, and the paper states they were 'tuned thoroughly by means of simulations.' No sensitivity analysis is provided for these parameters, and no results are shown for nearby operating points. Since the congestion detector's decisions and therefore the ARN+AFI behavior depend heavily on these thresholds, the reader cannot judge whether the observed improvement is robust or an artifact of a carefully chosen configuration. Please add a sensitivity study or at least report the results for a plausible range of parameter values.
minor comments (5)
  1. [Figure 8 caption] The caption refers to 'configuration #1 in Table 2,' but the network configurations are listed in Table 3; Table 2 is the ARN table entry fields. Please correct the cross-reference.
  2. [Section 4.1, ARN mechanism paragraph] The text 'CITAR [Hipineb17], [NVIDIA-quantum]' is garbled and does not match the reference list; it should be replaced with the actual citations for the Mellanox/NVIDIA documentation (likely [31,32]).
  3. [Section 3.4 and Pseudo-code 4] The surrounding text refers to 'Algorithm 2' when the intended target is Pseudo-code 2; please use consistent terminology.
  4. [Section 3.2 vs. Section 4.1] Section 3.2 says the switch architecture defines at least two VCs per input port (one regular and the AFC), but Section 4.1 states that all non-AFI techniques use only 1 VC. Please clarify whether the 1Q configuration omits the AFC or uses a single VC with a different logical role.
  5. [Section 2.2, paragraph on fat-tree ARN consumption] The statement that 'if there is a congestion situation in a downward path, packets following that path must be adapted at some point of the upward phase' is stated as a general rule but is only true when the congestion root is in the downward phase and an alternative upward path exists; the condition should be stated more precisely.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ARN+AFI is an empirical simulation study whose conclusions are generated by the simulator model, not forced by construction, fitted parameters, or load-bearing self-citation.

full rationale

The paper's central claim is an empirical one: in the SAURON simulator, ARN+AFI recovers network efficiency faster than D-mod-K, oblivious routing, threshold-based adaptive routing, AFI, and the modeled plain ARN mechanism under incast and MPI trace workloads. No equation in the paper defines the measured efficiency in terms of the proposed mechanism's own outputs. The ARN+AFI design does include a by-construction property—adapted packets are placed in the AFC and therefore cannot occupy the regular VCs used by non-adapted flows—but the evaluation still measures dynamic behavior (reaction time, efficiency recovery, application runtime) that is not a logical consequence of that queue assignment alone, and it compares against external baselines (D-mod-K, oblivious routing, PTRANS, Inception-V3 traces). Parameter values (HCDTh, LCDTh, FCTh, CRT) are disclosed as tuned by simulation, so the results are conditional on that tuning rather than disguised predictions. Self-citations to [26] and [27] provide background and the AFI building block, and [43] is the simulator; none of these is used in place of an argument that the simulated benefit must occur. The weakest assumption—that real InfiniBand switches implement the ARN forwarding and consumption semantics modeled in Pseudo-codes 1 through 4—is a correctness and falsifiability risk about an external premise, not a circularity within the paper's derivation chain. Therefore no circular step is exhibited, and the appropriate score is 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central simulator results rest on five tuned thresholds, several structural assumptions about fat-tree routing and switch architecture, and the untested assumption that ARN hardware behaves as specified in the Mellanox patent. No new physical entities are introduced: the AFC and the adapted bit are inherited from prior AFI work, and ARN tables and messages are inherited from Mellanox.

free parameters (5)
  • HCDTh = 81% of buffer occupancy
    High congestion detection threshold; explicitly tuned by simulations to trigger congestion-root declaration.
  • LCDTh = 63% of buffer occupancy
    Low congestion detection threshold; tuned based on Gran et al. experiments, used to disable congestion roots.
  • FCTh = 78% of free credits in next buffer
    Free-credit threshold that distinguishes a congestion root from a tree branch; tuned assuming several packets in flight.
  • CRT = 5 ms
    Congestion root timer; tuned for the fastest detector reaction when congestion trees are in steady state.
  • Adaptive-Th trigger threshold = 75% buffer occupancy
    Used only for the Adaptive-Th baseline; manually tuned so the baseline adapts only under strong and lasting congestion. Not part of ARN+AFI but affects the fairness of the comparison.
assumptions (5)
  • domain assumption In a fat-tree, the downward phase has a unique path to each destination, so ARNs can be consumed only in the upward phase at switches one or more stages below the congestion root.
    Used in Section 2.2 and Pseudo-code 1 to decide where ARNs are consumed and to compute congestionRootInfo from the switch stage.
  • domain assumption Switches are input-queued with per-VC credit-based flow control and virtual output queues, and VOQs are not flow-controlled and can hog buffer space.
    Section 3.2: the whole congestion detector and AFC isolation design depends on this switch architecture.
  • domain assumption Adaptive routing notifications behave as described in the Mellanox patent and vendor documentation: they can be generated at a congested port, forwarded upstream, consumed at lower stages, and managed in an ARN table with timers.
    Sections 2.2 and 3.4 rely on this behavior; no independent hardware measurements are cited.
  • ad hoc to paper The packet at the head of the VOQ that crosses the HCDTh is the one responsible for the congestion situation.
    Section 3.3 states this directly. It is the basis for filling the ARN destination field and is not validated against actual congesting-flow membership.
  • ad hoc to paper A congested output port is a congestion root if free credits in the next switch exceed FCTh, and a branch if they do not.
    Section 3.3 introduces this heuristic, citing only the InfiniBand specification criterion for a related but not identical check.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Congestion Management in High-Performance Interconnection Networks Using Adaptive Routing Notifications." pith.science (2026). https://pith.science/paper/F5YQKEY4

@misc{pith2026250200616,
  author       = {Pith},
  title        = {Pith review of: Congestion Management in High-Performance Interconnection Networks Using Adaptive Routing Notifications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F5YQKEY4}},
  note         = {Machine review of arXiv:2502.00616}
}
read the original abstract

The interconnection network is a crucial subsystem in High-Performance Computing clusters and Data-centers, guaranteeing high bandwidth and low latency to the applications' communication operations. Unfortunately, congestion situations may spoil network performance unless the network design applies specific countermeasures. Adaptive routing algorithms are a traditional approach to dealing with congestion since they provide traffic flows with alternative routes that bypass congested areas. However, adaptive routing decisions at switches are typically based on local information without a global network traffic perspective, leading to congestion spreading throughout the network beyond the original congested areas. In this paper, we propose a new efficient congestion management strategy that leverages adaptive routing notifications currently available in some interconnect technologies and efficiently isolates the congesting flows in reserved spaces at switch buffers. The experiment results based on simulations of realistic traffic scenarios show that our proposal removes the congestion impact.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 35 canonical work pages

  1. [40]

    US20140211631A1, Apr 2015

    Haramaty, Z., Zahavi, E., Gabbay, F., Crupnicoff, D., Marelli, A., Bloch, G.: Adaptive Routing Using Inter-switch Notifications. US20140211631A1, Apr 2015

    work page 2015

  2. [1]

    IEEE Transactions on Computers C-34(10), 892–901 (1985)

    Leiserson, C.E.: Fat-trees: Universal networks for hardware-efficient supercomputing. IEEE Transactions on Computers C-34(10), 892–901 (1985). https://doi.org/10.1109/TC.1985.6312192

    work page arXiv 1985

  3. [2]

    In: 21th International Parallel and Distributed Processing Symposium (IPDPS) 2007, Proceed- ings, 26-30 March 2007, Long Beach, California, USA, pp

    Requena, C.G., Villam´ on, F.G., G´ omez, M.E., L´ opez, P., Duato, J.: Deter- ministic versus Adaptive Routing in Fat-Trees. In: 21th International Parallel and Distributed Processing Symposium (IPDPS) 2007, Proceed- ings, 26-30 March 2007, Long Beach, California, USA, pp. 1–8. IEEE, ??? (2007). https://doi.org/10.1109/IPDPS.2007.370482

    work page arXiv 2007

  4. [3]

    Journal of CCPE 22(2), 217–231 (2010)

    Zahavi, E., Johnson, G., Kerbyson, D.J., Lang, M.: Optimized InfiniBandTM fat-tree routing for shift all-to-all communication patterns. Journal of CCPE 22(2), 217–231 (2010)

    work page 2010

  5. [4]

    In: 2009 IEEE International Conference on Cluster Computing and Workshops, pp

    Rodriguez, G., Minkenberg, C., Beivide, R., Luijten, R.P., Labarta, J., Valero, M.: Oblivious routing schemes in extended generalized fat tree Springer Nature 2021 LATEX template 30 Congestion Management in Interconnects using Adaptive Routing networks. In: 2009 IEEE International Conference on Cluster Computing and Workshops, pp. 1–8 (2009). https://doi....

    work page doi:10.1109/clustr.2009 2009

  6. [5]

    IEEE Journal on Selected Areas in Communications 32(1), 88–101 (2014)

    Zahavi, E., Keslassy, I., Kolodny, A.: Distributed adaptive routing con- vergence to non-blocking DCN routing assignments. IEEE Journal on Selected Areas in Communications 32(1), 88–101 (2014). https://doi.org/ 10.1109/JSAC.2014.140109

    work page arXiv 2014

  7. [6]

    In: 16th Annual IEEE Symposium on High Per- formance Interconnects (HOTI 2008), 26-28 August 2008, Stanford, CA, USA, pp

    Geoffray, P., Hoefler, T.: Adaptive routing strategies for modern high performance networks. In: 16th Annual IEEE Symposium on High Per- formance Interconnects (HOTI 2008), 26-28 August 2008, Stanford, CA, USA, pp. 165–172. IEEE Computer Society, ??? (2008). https://doi.org/ 10.1109/HOTI.2008.21. http://dx.doi.org/10.1109/HOTI.2008.21

    work page doi:10.1109/hoti.2008.21 2008

  8. [7]

    In: Proceedings of the ACM/IEEE SC2006 Conference on High Performance Networking and Computing, November 11-17, 2006, Tampa, FL, USA, p

    Kim, J., Dally, W.J., Abts, D.: Interconnect routing and scheduling - adaptive routing in high-radix clos network. In: Proceedings of the ACM/IEEE SC2006 Conference on High Performance Networking and Computing, November 11-17, 2006, Tampa, FL, USA, p. 92. ACM Press, ??? (2006). https://doi.org/10.1145/1188455.1188552. http://doi. acm.org/10.1145/1188455.1188552

    work page arXiv 2006

Show all 50 references
  1. [8]

    In: 2008 IEEE 14th International Sympo- sium on High Performance Computer Architecture, pp

    Gratz, P., Grot, B., Keckler, S.W.: Regional congestion awareness for load balance in networks-on-chip. In: 2008 IEEE 14th International Sympo- sium on High Performance Computer Architecture, pp. 203–214 (2008). https://doi.org/10.1109/HPCA.2008.4658640

    2008

  2. [9]

    In: 2011 38th Annual International Symposium on Computer Architecture (ISCA), pp

    Ma, S., Jerger, N.E., Wang, Z.: Dbar: An efficient routing algorithm to support multiple concurrent applications in networks-on-chip. In: 2011 38th Annual International Symposium on Computer Architecture (ISCA), pp. 413–424 (2011)

    2011

  3. [10]

    RFC 2992, 1–8 (2000)

    Hopps, C.E.: Analysis of an equal-cost multi-path algorithm. RFC 2992, 1–8 (2000). https://doi.org/10.17487/RFC2992

    2000 doi

  4. [11]

    In: Uhlig, S., Maennel, O., Karp, B., Padhye, J

    He, K., Rozner, E., Agarwal, K., Felter, W., Carter, J.B., Akella, A.: Presto: Edge-based load balancing for fast datacenter networks. In: Uhlig, S., Maennel, O., Karp, B., Padhye, J. (eds.) Proceedings of the 2015 ACM Conference on Special Interest Group on Data Communication...

    2015

  5. [12]

    Ghorbani, S., Yang, Z., Godfrey, P.B., Ganjali, Y., Firoozshahian, A.: DRILL: micro load balancing for low-latency data center networks. In: Proceedings of the Conference of the ACM Special Interest Group on Data Communication, SIGCOMM 2017, Los Angeles, CA, USA, August 21-25,...

    2017

  6. [13]

    IEEE Transactions on Network and Service Management 15(4), 1706–1719 (2018)

    Wang, S., Luo, J., Wong, W.S.: Improved power of two choices for fat-tree routing. IEEE Transactions on Network and Service Management 15(4), 1706–1719 (2018). https://doi.org/10.1109/TNSM.2018.2865543

    2018

  7. [14]

    In: Damkroger, T., Dongarra, J

    Besta, M., Hoefler, T.: Slim Fly: A Cost Effective Low-Diameter Network Topology. In: Damkroger, T., Dongarra, J. (eds.) International Conference for High Performance Computing, Networking, Storage and Analysis, SC 2014, New Orleans, LA, USA, November 16-21, 2014, pp. 348–359....

    2014 doi

  8. [15]

    In: Gribble, S.D., Katabi, D

    Singla, A., Hong, C., Popa, L., Godfrey, P.B.: Jellyfish: Networking data centers randomly. In: Gribble, S.D., Katabi, D. (eds.) Proceedings of the 9th USENIX Symposium on Networked Systems Design and Implemen- tation, NSDI 2012, San Jose, CA, USA, April 25-27, 2012, pp. 225–2...

    2012

  9. [16]

    In: 35th International Symposium on Computer Architecture (ISCA) 2008, June 21-25, 2008, Beijing, China, pp

    Kim, J., Dally, W.J., Scott, S., Abts, D.: Technology-Driven, Highly- Scalable Dragonfly Topology. In: 35th International Symposium on Computer Architecture (ISCA) 2008, June 21-25, 2008, Beijing, China, pp. 77–88. IEEE Computer Society, ??? (2008). https://doi.org/10.1109/ IS...

    2008

  10. [17]

    Valiant, L.G.: A Scheme for Fast Parallel Communication. SIAM J. Comput. 11(2), 350–361 (1982). https://doi.org/10.1137/0211027

    1982 doi

  11. [18]

    In: 36th International Symposium on Computer Architecture (ISCA 2009), June 20-24, 2009, Austin, TX, USA, pp

    Jiang, N., Kim, J., Dally, W.J.: Indirect adaptive routing on large scale interconnection networks. In: 36th International Symposium on Computer Architecture (ISCA 2009), June 20-24, 2009, Austin, TX, USA, pp. 220– 231 (2009). https://doi.org/10.1145/1555754.1555783

    2009

  12. [19]

    In: The 21st IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing, CCGrid, May 10-13, 2021, Melbourne, Victoria, Australia, p

    Newaz, M.N., Mollah, M.A., Faizian, P., Tong, Z.: Improving adap- tive routing performance on large scale megafly topology. In: The 21st IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing, CCGrid, May 10-13, 2021, Melbourne, Victoria, Australia, p

    2021

  13. [20]

    IEEE Trans

    Besta, M., Domke, J., Schneider, M., Konieczny, M., Girolamo, S.D., Schneider, T., Singla, A., Hoefler, T.: High-performance routing with multipathing and path diversity in ethernet and HPC networks. IEEE Trans. Parallel Distributed Syst. 32(4), 943–959 (2021). https://doi.org...

    2021

  14. [21]

    IEEE/ACM, ??? (2021)

    2021

  15. [22]

    In: High Performance Embedded Architectures and Com- pilers, First International Conference, HiPEAC 2005, Barcelona, Spain, November 17-18, 2005, Proceedings, pp

    Garc ´ ıa, P.J., Flich, J., Duato, J., Johnson, I., Quiles, F.J., Naven, F.: Dynamic evolution of congestion trees: Analysis and impact on switch architecture. In: High Performance Embedded Architectures and Com- pilers, First International Conference, HiPEAC 2005, Barcelona, ...

    2005 doi

  16. [23]

    IEEE SA Industry Connections–IEEE 802 Nendica Report: Intelligent Lossless Data Center Networks, 1–44 (2021)

    Guo, L., Congdon, P.: Ieee 802 nendica report: Intelligent lossless data cen- ter networks. IEEE SA Industry Connections–IEEE 802 Nendica Report: Intelligent Lossless Data Center Networks, 1–44 (2021)

    2021

  17. [24]

    In: 2007 IEEE International Conference on Communications, pp

    Yoshigoe, K.: Threshold-based exhaustive round-robin for the cicq switch with virtual crosspoint queues. In: 2007 IEEE International Conference on Communications, pp. 6325–6329 (2007). https://doi.org/10.1109/ICC. 2007.1047

    2007

  18. [25]

    IEEE Trans

    Karol, M.J., Hluchyj, M.G., Morgan, S.P.: Input versus output queueing on a space-division packet switch. IEEE Trans. Communications 35(12), 1347–1356 (1987). https://doi.org/10.1109/TCOM.1987.1096719

    1987

  19. [26]

    Rocher-Gonzalez, J., Escudero-Sahuquillo, J., Garc ´ ıa, P.J., Quiles, F.J., Mora, G.: Towards an efficient combination of adaptive routing and queu- ing schemes in fat-tree topologies. J. Parallel Distributed Comput. 147, 46–63 (2021). https://doi.org/10.1016/j.jpdc.2020.07.009

    2021 doi

  20. [27]

    Jurczyk, M., Schwederski, T.: Phenomenon of Higher Order Head-of- Line Blocking in Multistage Interconnection Networks under Nonuniform Traffic Patterns (1996)

    1996

  21. [28]

    Parallel and Distributed Systems, IEEE Trans- actions on 21(6), 739–753 (2010)

    Nachiondo, T., Flich, J., Duato, J.: Buffer Management Strategies to Reduce HoL Blocking. Parallel and Distributed Systems, IEEE Trans- actions on 21(6), 739–753 (2010). https://doi.org/10.1109/TPDS.2009. 63

    2010 doi

  22. [29]

    In: 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGRID 2019, Larnaca, Cyprus, May 14-17, 2019, pp

    Rocher-Gonzalez, J., Escudero-Sahuquillo, J., Garc ´ ıa, P.J., Flor, F.J.Q., Mora, G.: Efficient congestion management for high-speed interconnects using adaptive routing. In: 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGRID 2019, Larnaca, Cyp...

    2019

  23. [30]

    Escudero-Sahuquillo, J., Garc ´ ıa, P.J., Quiles, F.J., Reinemo, S., Skeie, T., Lysne, O., Duato, J.: A new proposal to deal with congestion in InfiniBand-based fat-trees. J. Parallel Distrib. Comput. 74(1), 1802–1819 (2014). https://doi.org/10.1016/j.jpdc.2013.09.002

    2014 doi

  24. [31]

    In: 25th IEEE International Symposium on Parallel and Distributed Processing, IPDPS 2011, Anchorage, Alaska, USA, 16-20 May, 2011 - Con- ference Proceedings, pp

    Guay, W.L., Bogdanski, B., Reinemo, S., Lysne, O., Skeie, T.: vFtree - A Fat-Tree Routing Algorithm Using Virtual Lanes to Alleviate Conges- tion. In: 25th IEEE International Symposium on Parallel and Distributed Processing, IPDPS 2011, Anchorage, Alaska, USA, 16-20 May, 2011 ...

    2011 doi

  25. [32]

    https://support.mellanox.com/s/article/ How-To-Configure-Adaptive-Routing-and-Self-Healing-Networking-New Accessed 2021-08-05

    Mellanox: How To Configure Adaptive Routing and SHIELD (New). https://support.mellanox.com/s/article/ How-To-Configure-Adaptive-Routing-and-Self-Healing-Networking-New Accessed 2021-08-05

    2021

  26. [33]

    https://network.nvidia.com/sites/default/files/doc-2020/ pb-quantum-hdr-switch-silicon.pdf Accessed 2020-09-15

    Mellanox: NVIDIA MELLANOX QUANTUM -PRODUCT BRIEF. https://network.nvidia.com/sites/default/files/doc-2020/ pb-quantum-hdr-switch-silicon.pdf Accessed 2020-09-15

    2020

  27. [34]

    In: Proceedings of the 2009 IEEE International Conference on Cluster Computing, August 31 - September 4, 2009, New Orleans, Louisiana, USA, pp

    Rodr ´ ıguez, G., Minkenberg, C., Beivide, R., Luijten, R.P., Labarta, J., Valero, M.: Oblivious routing schemes in extended generalized fat tree networks. In: Proceedings of the 2009 IEEE International Conference on Cluster Computing, August 31 - September 4, 2009, New Orlean...

    2009

  28. [35]

    In: 25th IEEE International Sympo- sium on Parallel and Distributed Processing, IPDPS 2011, Anchorage, Alaska, USA, 16-20 May 2011 - Workshop Proceedings, pp

    Zahavi, E.: Fat-trees routing and node ordering providing contention free traffic for MPI global collectives. In: 25th IEEE International Sympo- sium on Parallel and Distributed Processing, IPDPS 2011, Anchorage, Alaska, USA, 16-20 May 2011 - Workshop Proceedings, pp. 761–770....

    2011 doi

  29. [36]

    Intel ® omni-path fabric suite fabric manager. (2015). https://www.intel. com/content/dam/support/us/en/documents/network/omni-adptr/sb/ Intel OP FabricSuite Fabric Manager UG H76468 v1 0.pdf

    2015

  30. [37]

    Concurrency and Computation: Practice and Experience 22(2), 217–231 (2010)

    Zahavi, E., Johnson, G., Kerbyson, D.J., Lang, M.: Optimized InfiniBand fat-tree routing for shift all-to-all communication patterns. Concurrency and Computation: Practice and Experience 22(2), 217–231 (2010). https: //doi.org/10.1002/cpe.1527

    2010 doi

  31. [38]

    1–14 (2020)

    De Sensi, D., Di Girolamo, S., McMahon, K., Roweth, D., Hoefler, T.: An in-depth analysis of the slingshot interconnect, pp. 1–14 (2020). https: //doi.org/10.1109/SC41405.2020.00039

    2020 arXiv

  32. [39]

    The Journal of Supercomputing 72 (2016)

    Vign´ eras, P., Quintin, J.-N.: The bxi routing architecture for exascale supercomputer. The Journal of Supercomputing 72 (2016). https://doi. org/10.1007/s11227-016-1755-2

    2016 doi

  33. [41]

    InfiniBand Trade Association.: InfiniBandTM Architecture Specification Springer Nature 2021 LATEX template 34 Congestion Management in Interconnects using Adaptive Routing Volume 1 - Release 1.3 (2015)

    2015

  34. [42]

    Journal of Parallel and Distributed Computing 112, 35–52 (2018)

    Escudero-Sahuquillo, J., Garcia, P.J., Quiles, F.J., Maglione-Mathey, G., Duato, J.: Feasible enhancements to congestion control in infiniband- based networks. Journal of Parallel and Distributed Computing 112, 35–52 (2018). https://doi.org/10.1016/j.jpdc.2017.09.008

    2018 doi

  35. [43]

    In: 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2011, Newport Beach, CA, USA, May 23-26, 2011, pp

    Gran, E.G., Zahavi, E., Reinemo, S., Skeie, T., Shainer, G., Lysne, O.: On the relation between congestion control, switch arbitration and fairness. In: 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2011, Newport Beach, CA, USA, May 23-26, ...

    2011 doi

  36. [44]

    OpenSim Ltd: OMNeT++ Discrete Event Simulator

  37. [45]

    In: 21st Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2013, Belfast, United Kingdom, Febru- ary 27 - March 1, 2013, pp

    Yebenes, P., Escudero-Sahuquillo, J., Garc ´ ıa, P.J., Quiles, F.J.: Towards modeling interconnection networks of exascale systems with omnet++. In: 21st Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, PDP 2013, Belfast, United Kingdo...

    2013 doi

  38. [46]

    The Journal of Super- computing 72(12), 4601–4628 (2016)

    Andujar, F.J., Villar, J.A., Alfaro, F.J., S´ anchez, J.L., Escudero- Sahuquillo, J.: An open-source family of tools to reproduce mpi-based workloads in interconnection network simulators. The Journal of Super- computing 72(12), 4601–4628 (2016)

    2016

  39. [47]

    In: 2011 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, pp

    Gran, E.G., Zahavi, E., Reinemo, S.-A., Skeie, T., Shainer, G., Lysne, O.: On the relation between congestion control, switch arbitration and fairness. In: 2011 11th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, pp. 342–351 (2011). https://doi.org/10.1...

    2011

  40. [48]

    In: 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV, USA, June 27-30, 2016, pp

    Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., Wojna, Z.: Rethinking the inception architecture for computer vision. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV, USA, June 27-30, 2016, pp. 2818–2826. IEEE Computer Society, ...

    2016 doi

  41. [49]

    http://icl.cs.utk.edu/hpcc/

    The HPCC Benchmark. http://icl.cs.utk.edu/hpcc/. http://icl.cs.utk. edu/hpcc/ Accessed 2016-12-19

    2016

  42. [478]

    https://doi.org/10.1145/2785956.2787507

    ACM, ??? (2015). https://doi.org/10.1145/2785956.2787507. https: //doi.org/10.1145/2785956.2787507

    2015

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.