Pith. sign in

REVIEW 2 major objections 5 minor 39 references

Taurus: Accelerating Out-of-Core Graph Neural Network Inference on Billion-Scale Graphs

T0 review · 2 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Pith's one-line read Out-of-core GNN inference can be made sequential: broadcast beats gather on billion-scale graphs.

desk verdict Worth engaging seriously, but the IGB-Full headline numbers (25x/140x) are extrapolations from a baseline that timed out, so treat the upper end of the range as indicative until the artifact ships. read the letter →

arxiv 2607.17374 v1 pith:3L7AKCY2 submitted 2026-07-19 cs.DC cs.LG

classification cs.DCcs.LG
keywords graphneuralnetworksout-of-coreinferencebillion-scalegraphssource-centricbroadcastsequentialSSDscansreorderingevictionpolicysingle-machinesystems
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

Taurus claims that graph-neural-network inference on graphs too large for RAM can be made practical on a single workstation by reversing the direction of message passing. Instead of each destination vertex gathering neighbor embeddings from disk—which produces random, repeated, block-amplified reads—each source vertex broadcasts its embedding once in vertex-ID order, turning the workload into sequential SSD scans. The paper shows that the resulting bottleneck, partially aggregated destination states, can be controlled with a tiered GPU–RAM–SSD store, a topology-aware reordering that shortens the interval between a vertex's first and last incoming message, and an eviction policy that spills the states closest to completion. On graphs with up to 269 million vertices, 4 billion edges, and 514 GiB of features, it reports completing exact and fanout-sampled inference in under 30 minutes on the largest graph, with 7–25× speedups over the strongest layer-wise baseline and 40–140× over vertex-wise baselines. The significance, if the claims hold, is that billion-scale GNN inference no longer requires a distributed cluster; a 128 GiB workstation with one GPU and a 2 TiB SSD suffices.

What carries the argument

The source-centric broadcast execution model, backed by a three-tier GPU–RAM–SSD partial-state store. Messages flow along out-edges in CSR vertex order, so each embedding is read once; destination partial aggregates live in a pinned GPU store for hubs, a bounded hot store in RAM, and an SSD cold store with merge-on-read run files. Two controls carry the performance argument: topology-aware reordering minimizes the cumulative active-state span C(π) via a differentiable dispersion objective J, and the eviction policy spills the k states with the fewest pending messages, minimizing eviction–reload cycles.

What would settle it

Engineer a graph with many destinations whose in-neighbors are deliberately spread across distant rank intervals after reordering (e.g., a planted-community graph with long-range edges), run Taurus and a layer-wise gather baseline, and measure inference time and reload counts against C(π) and J. If J converges to an ordering that does not reduce span, or if reloads and runtime stop tracking C(π) while sequential reads remain, the central claim is falsified; alternatively, a simple comparison where Taurus's speedup over the layer-wise baseline drops below the reported range on such a topology w

Watch

Extended reading notes

Core claim

The paper's central claim is that layer-wise GNN inference can be reformulated as source-centric broadcasts, replacing repeated random gathers with sequential disk access. Taurus implements this by streaming each vertex's embedding once per pass from SSD, buffering partial aggregation states for destinations in a GPU-resident store for high-degree vertices, a hot store in RAM, and an SSD cold store, and writing completed embeddings as sorted run files that are merged on read. To keep partial-state lifetimes short, it reorders vertices by minimizing the total span—the sum over destinations of the distance between the first and last incoming source—using an iterative in-neighborhood dispersion

Load-bearing premise

The performance gains rest on the empirical chain—validated on only two graphs and two starting orderings—that total span C(π) predicts inference time and cold-store reloads, and that the differentiable dispersion objective J tracks C(π); if a graph topology breaks that correlation, the sequential-scan advantage erodes toward gather-based baselines.

Editorial extensions

If this is right

  • If correct, exact full-neighborhood inference on a graph with 514 GiB of features finishes in under 30 minutes on a single workstation with 128 GiB RAM, one GPU, and a 2 TiB SSD, making periodic refresh of predictions on evolving graphs practical within hours.
  • Sequential broadcast cuts disk traffic by large factors: the paper reports the vertex-wise baselines read up to roughly 50× and 108× more data than Taurus in the evaluated workloads.
  • Alternative GNN operators—GraphSAGE, GAT, GIN—can run through additional sequential passes rather than random gathers, preserving semantics without materializing edge features, at a modest runtime cost.
  • Topology-aware reordering and pending-message eviction reduce cold-store reloads by up to two orders of magnitude, and Taurus at a 60 GiB hot store matches the performance of other orderings at roughly 90 GiB, decoupling inference speed from RAM scale.

Reading between the lines

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

  • The broadcast-plus-partial-state pattern is general: any vertex-centric iterative computation with commutative aggregation (e.g., PageRank-style scoring or label propagation) could adopt the same sequential-scan, tiered-buffer design for out-of-core graphs, not just GNNs.
  • The reordering objective's dependence on static topology means graphs with rapidly changing edge structure could invalidate a cached ordering; an incremental reordering that updates positions under edge insertions would be a natural testable extension.
  • If the span-to-runtime correlation holds across topologies, the system's gains should transfer to other power-law and community-structured graphs; a hybrid that gathers over long-range edges only while broadcasting locally could extend the design to graphs where reordering cannot compress span.
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

2 major / 5 minor

Summary. Taurus is a single-machine, out-of-core inference engine for GNNs on graphs whose feature/embedding stores exceed RAM. The core idea is to replace the standard destination-centric gather (each vertex pulls its neighbors' embeddings) with a source-centric broadcast that streams each source embedding once in vertex order, using a CSR out-edge layout and a tiered GPU–RAM–SSD aggregation hierarchy. Partial aggregation states are managed under a bounded hot-store budget, with a GPU-resident store for high-degree vertices, a pending-message-count eviction policy, and run-file output to avoid random writes. A topology-aware reordering procedure minimizes a differentiable neighborhood-dispersion objective, which is empirically shown to track total vertex span and hence reload/I/O cost. The paper evaluates exact and fanout-sampled inference on five public datasets (up to 269M vertices, 4B edges, 514 GiB features), reporting 7–25× speedups over the layer-wise DGI baseline and 40–140× over vertex-wise Ginex/DGL baselines, with ablations of reordering, eviction, hot-store/GPU-store capacity, and a comparison to the authors' prior ATLAS system.

Significance. If the reported numbers hold, Taurus makes a strong case that single-node OOC inference can outperform distributed or disk-based gathering systems at billion scale, and the broadcast formulation is a conceptually clean alternative to gather-based execution. The evaluation is generally careful: public datasets, page-cache clearing, a numerical cross-check of exact inference against in-memory DGL (mean per-vertex max abs error 8e-5, mean rel error 2.8e-6), and an honest negative result for in-memory SAGE/GAT on PA (0.6× of DGI). The ablations isolate the effect of reordering, eviction, and memory budgets, which is valuable. The main caveat is that the highest headline speedups on IGB-Full depend on extrapolated baseline runtimes rather than direct measurement.

major comments (2)
  1. [§4.2, §4.3, Fig. 6] The headline upper bounds on IGB-Full (25× over DGI, 140×/96× over Ginex/DGL) are not measured runtimes. §4.2 caps all runs at 4h and linearly extrapolates incomplete runs from completed vertex ranges/chunks; §4.3 says DGI's first layer on IF exceeded the cap, so its total is an extrapolation plus later layers with dummy inputs. No completed fraction, no extrapolation error, and no validation of the linearity assumption on a dataset where DGI completes is reported. A 30–40% error in the first-layer estimate would move 25× into the low-20s or high-teens. Please either complete the DGI first layer on IF (or a representative subgraph) or validate the extrapolation procedure against full runs on FS/MA/IL, and report the completed fraction and error bars. The same caveat applies to GX/DG bars on IF/MA.
  2. [§3.8, Fig. 5] The J-to-C(π)-to-time chain is validated only on FS and IL (R²=0.83–0.99). Yet the converged TA ordering is applied to MA and IF and to SAGE/GAT models in Fig. 6; if the correlation breaks on these topologies, the reordering benefit (and the magnitude of the speedup) would shrink. Please report C(π) and runtime for the converged ordering on MA and IF, or explicitly restrict the reordering claims to the validated datasets.
minor comments (5)
  1. [Abstract and §4.3] The advertised speedup ranges are for fanout-sampled inference only. Exact full-neighborhood inference is benchmarked only against ATLAS and not against DGI/Ginex/DGL; please state this qualification in the abstract or at the start of §4.3.
  2. [§3.9] The fanout-sampled edge-mask construction is described in one sentence. Since the headline evaluation is sampled, specify how the mask is generated, stored, and applied during the broadcast pass, and its memory/disk cost.
  3. [§3.8, Algorithm 1] The formula for r*(w) divides by the sum of 1/d⁻(v) over v ∈ N⁺(w); for vertices with no out-neighbors the denominator is zero. Specify the fallback (keep previous position, assign a constant, etc.).
  4. [§4.1] The numerical cross-check against in-memory DGL is only for exact inference on PA. A sampled-inference cross-check, even on a smaller graph, would increase confidence in the sampled mode used for the headline comparison.
  5. [General] Use R² notation consistently; the annotation in Fig. 4 ('E VO GC PH SN S') appears garbled; 'MAG-Cites' is once written as 'MA G-Cites'; and no artifact/code availability statement is included.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; central speedup rests on external baselines and public datasets; minor self-citations (ATLAS) are ablative/provenance, not load-bearing.

full rationale

The paper's central claim — that Taurus outperforms DGI by 7–25× and Ginex/DGL by 40–140× on billion-scale out-of-core graphs — is an empirical comparison against external systems (DGI, Ginex, DGL) on public datasets (OGB, SNAP, IGB). No target result is fed into Taurus to produce its output; the broadcast execution model is shown to be semantically equivalent to gather by construction, and the measured speedups are direct runtime comparisons. The topology-aware reordering objective J is validated by empirical correlations with C(π) and inference time (Fig. 5), and C(π) itself is defined independently from the runtime as cumulative active-state occupancy; this is a validation, not a self-definition. The only notable self-citation is ATLAS [1], which is acknowledged as prior work that introduced broadcast-based layer-wise inference and is used as an ablation baseline in Fig. 13. That self-citation is not load-bearing for the central speedup claim: the main comparisons are against DGI, Ginex, and DGL, and Taurus's design is described self-containedly in §3. The IGB-Full result uses extrapolated baseline runtimes for runs that exceeded the 4h cap, which is a measurement/correctness concern, not a circularity: the extrapolation is not an input to Taurus's derivation. No fitted parameter is renamed as a prediction, and no uniqueness theorem from the authors' prior work is invoked to forbid alternatives. Overall, the derivation chain is self-contained against external benchmarks; residual circularity is limited to minor self-citation that does not affect the central result.

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

Systems paper: the 'free parameters' are hand-chosen runtime budgets and evaluation constants (memory tiers, fanout, iteration count), not constants fitted to a derivation. The only scientific-type heuristic is the reordering objective: minimizing dispersion J is justified by empirical correlation with span C(π) and runtime on two datasets, so it is classified as ad-hoc-to-paper. No new physical entities are postulated; the GPU store, hot/cold stores, and unmerged run files are software constructs whose behavior is validated experimentally in §4.

free parameters (5)
  • hot store size (RAM budget for partial aggregation states) = 50 GiB (PA/MA/IL); 100 GiB (FS/IF)
    Hand-chosen per dataset (§4.2). Ablations (Figs. 10–11) show Taurus is flatter than baselines vs this budget, so headline numbers are not primarily driven by it.
  • GPU store size = 16 GiB
    VRAM dedicated to pinning high-in-degree hub states (§4.2). Fig. 12 shows 64–72% of benefit arrives within the first 8 GiB; 16 GiB is an operating-point choice.
  • chunk size / graduation buffer / queue size = 8 MiB / 256 MiB / 20
    Fixed pipeline constants (§4.2); not swept in the main evaluation.
  • reordering iterations = 5
    Chosen because 84–90% of span reduction occurs in 4–5 iterations (§4.4); all reported results use this fixed count.
  • sampling fanout = 10 per layer
    Evaluation constant for sampled inference (§4.3); justified as a middle point between understating message-propagation cost and exceeding baseline time budgets.
assumptions (5)
  • standard math Cumulative active-state occupancy equals total span: M = Σ_t A(t) = Σ_v L(v), with L(v) = max π(u) − min π(u) over in-neighbors of v.
    Counting identity in §3.8; exact for any ordering π and the foundation of the reordering objective.
  • domain assumption Every vertex aggregates a self message, so d⁻(v) ≥ 1 and the 1/d⁻(v) weighting in the dispersion objective J is well-defined.
    §3.8. Holds for GCN/GIN-style models with self-loops; SAGE/GAT are handled separately in §3.9, but exact-inference normalization on loop-free graphs is not discussed.
  • ad hoc to paper Minimizing the differentiable dispersion J empirically tracks minimizing span C(π), and C(π) tracks inference time and cold-store reloads.
    Fig. 5: R²=0.96–0.99 (J vs C) and R²=0.83–0.96 (C vs time), both on FS and IL only. This empirical chain carries the claimed 31–75% runtime reduction from reordering (§4.4).
  • domain assumption Out-of-core GNN inference is I/O-bound and the OS page cache cannot absorb repeated random gathers.
    §1, Fig. 1b: 568 GiB physical reads vs 68 GiB graph size on PA despite page cache; motivates the broadcast design and O_DIRECT choice.
  • domain assumption Power-law degree distributions make GPU pinning of the largest in-degree hubs capture a large message share.
    §3.5.1, Fig. 12b: a 16 GiB GPU store pins 6.4% (FS) / 4.2% (IL) of vertices that receive 51% / 39% of destination messages.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Taurus: Accelerating Out-of-Core Graph Neural Network Inference on Billion-Scale Graphs." pith.science (2026). https://pith.science/paper/3L7AKCY2

@misc{pith2026260717374,
  author       = {Pith},
  title        = {Pith review of: Taurus: Accelerating Out-of-Core Graph Neural Network Inference on Billion-Scale Graphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3L7AKCY2}},
  note         = {Machine review of arXiv:2607.17374}
}
abstract

Graph Neural Network (GNN) inference on billion-scale graphs is challenging due to the large memory footprint of features and embeddings and high disk I/O costs in out-of-core settings. Existing distributed GNN systems incur high communication times and infrastructure costs, while disk-based GNN systems are primarily tailored to training and experience massive wasted reads during inference on the entire graph. We present Taurus, a single-machine system for GNN inference on graphs that do not fit in RAM, supporting both \textit{exact} full-graph inference and fanout-sampled inference. To avoid random and repeated feature gathers, Taurus reformulates layer-wise inference as source-centric broadcasts over sequential SSD scans, backed by a pipelined GPU-CPU-SSD hierarchy, topology-aware reordering, pending-message eviction, and a GPU-resident store for high-degree vertices. It further uses non-buffered sequential reads and GPU-backed writes to reduce page-cache pollution, host-memory pressure, and write overheads. On out-of-core graphs with up to $269M$ vertices, $4B$ edges, and $514$ GiB of features, Taurus outperforms the strongest layer-wise baseline, DGI, by $7$-$25\times$, and vertex-wise baselines by $40$-$140\times$.

Figures

Figures reproduced from arXiv: 2607.17374 by the authors.

Figure 1
Figure 1. Challenges limiting out-of-core GNN inference on [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Inference time (left Y axis, bars) and speedup relative to Taurus (right Y axis, markers) for 2-layer GraphConv inference with fanout=10 on disk-resident graphs (topol￾ogy+features) reported for GineX, DGL, DGI, and TAurus (ours) on PApers and MAG-Cites using a 5090 GPU. Proposal To address these challenges, we leverage a key insight: layer-wise GNN inference can be reformulated as source-centric broadcasts, enablin… view at source ↗
Figure 3
Figure 3. Gather-based versus broadcast-based execution for one GNN layer. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Taurus Architecture: Sequential graph/embedding reads, tiered aggregation, GPU [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Topology-based reordering on Friendster (FS) and IGB-large (IL) with RCMK (RC) [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Sampled inference performance (fanout=10 per layer) of TAurus vs. GineX, DGL, and DGI for 2-layer GCN, SAGE, GIN and GAT on PA, FS, MA, IL and IF. Bars show total inference time (s, log scale); markers show speedup over Taurus. Hatched bars denote extrapolated runtimes…
Figure 7
Figure 7. Figure 7: Impact of Taurus reordering bootstrap and iterations showing end-to-end inference [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Impact of Taurus reordering on execution behavior. (a) E2E inference time ( [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Impact of Taurus eviction on execution behavior. (a) E2E inference time ( [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Impact of hot-store capacity on E2E inference time under different vertex orderings [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Impact of hot-store capacity on TA. SSD I/O time ( [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Impact of GPU-store capacity on Taurus. (a) E2E inference time ( [PITH_FULL_IMAGE:figures/full_fig_p018_12.png]
Figure 13
Figure 13. Figure 13: Taurus (TA) vs. ATLAS (AT). E2E time (bars, left Y axis), I/O time (hatched), and # reloads (circles, right Y axis) for varying HS (+GS) budgets on IL (blue, left) and FS (green, right). of ≈ 12 and 9 min (Figs. 12a, bars, left Y axis). Increasing the GPU store perman…
Figure 14
Figure 14. Figure 14: Resource util. for a 2-layer GCN on the FS dataset. (a) CPU ( [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 3 linked inside Pith

  1. [1]

    Atlas: Efficient out-of-core inference for billion-scale graph neural networks,

    P. Naman and Y. Simmhan, “Atlas: Efficient out-of-core inference for billion-scale graph neural networks,” inACM Symposium on High-Performance Parallel and Distributed Computing (HPDC), 2026

  2. [2]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” inInternational Conference on Learning Representations (ICLR), 2017

  3. [3]

    Inductive representation learning on large graphs,

    W. L. Hamilton, R. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” inInternational Conference on Neural Information Processing Systems (NIPS), 2017

  4. [4]

    Enhancing graph neural network-based fraud detectors against camouflaged fraudsters,

    Y. Dou, Z. Liu, L. Sun, Y. Deng, H. Peng, and P. S. Yu, “Enhancing graph neural network-based fraud detectors against camouflaged fraudsters,” inACM International Conference on Information & Knowledge Management (CIKM), 2020

  5. [5]

    Billion-scale fintech analytics: Scalable data management and anomaly detection at npci,

    B. Dasari, T. S. Dhiraj, G. Jambhrunkar, T. Kailasam, C. Vikram, S. Singla, P. Na- man, and Y. Simmhan, “Billion-scale fintech analytics: Scalable data management and anomaly detection at npci,” inIEEE International Conference on Data Engineering (ICDE), 2026

  6. [6]

    Eta prediction with graph neural networks in google maps,

    A. Derrow-Pinion, J. She, D. Wong, O. Lange, T. Hester, L. Perez, M. Nunkesser, S. Lee, X. Guo, B. Wiltshireet al., “Eta prediction with graph neural networks in google maps,” inACM International Conference on Information & Knowledge Management (CIKM), 2021

  7. [7]

    Scaling real-time traffic analytics on edge- cloud fabrics for city-scale camera networks,

    A. Sharma, P. Naman, R. Banerjee, P. Pansari, S. Gawali, M. Arya, S. Chandra, A. Josephraj, R. Ramesh, P. Rathoreet al., “Scaling real-time traffic analytics on edge- cloud fabrics for city-scale camera networks,” inTCSC SCALE Challenge, IEEE CC- GRID Workshops, 2026

  8. [8]

    Aligraph: A comprehensive graph neural network platform,

    H. Yang, “Aligraph: A comprehensive graph neural network platform,” inACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2019

Show all 39 references
  1. [9]

    Inkstream: Instantaneous gnn inference on dynamic graphs via incremental update,

    D. Wu, Z. Li, and T. Mitra, “Inkstream: Instantaneous gnn inference on dynamic graphs via incremental update,” inIEEE International Parallel and Distributed Processing Sym- posium (IPDPS), 2025

  2. [10]

    Ripple: Scalable incremental gnn inferencing on large streaming graphs,

    P. Naman and Y. Simmhan, “Ripple: Scalable incremental gnn inferencing on large streaming graphs,” inIEEE International Conference on Distributed Computing Systems (ICDCS), 2025

  3. [11]

    Igb: Ad- dressing the gaps in labeling, features, heterogeneity, and size of public graph datasets for deep learning research,

    A. Khatua, V. S. Mailthody, B. Taleka, T. Ma, X. Song, and W.-m. Hwu, “Igb: Ad- dressing the gaps in labeling, features, heterogeneity, and size of public graph datasets for deep learning research,” inACM SIGKDD International Conference on Knowledge Discovery and Data Mining (...

  4. [12]

    Communication-efficient graph neural networks with probabilistic neighborhood expan- sionanalysisandcaching,

    T. Kaler, A. Iliopoulos, P. Murzynowski, T. Schardl, C. E. Leiserson, and J. Chen, “Communication-efficient graph neural networks with probabilistic neighborhood expan- sionanalysisandcaching,”Proceedings of Machine Learning and Systems (MLSys), 2023

  5. [13]

    Deal: distributed end-to-end gnn inference for all nodes,

    S. Chen, X. Song, V. Theodore, and H. Liu, “Deal: distributed end-to-end gnn inference for all nodes,”arXiv preprint arXiv:2503.02960, 2025

  6. [14]

    Diskgnn: Bridging i/o efficiency and model accuracy for out-of-core gnn training,

    R. Liu, Y. Wang, X. Yan, H. Jiang, Z. Cai, M. Wang, B. Tang, and J. Li, “Diskgnn: Bridging i/o efficiency and model accuracy for out-of-core gnn training,” inProceedings of the ACM on Management of Data (SIGMOD), 2025

  7. [15]

    Caliex: A disk-based large-scale gnn training system with joint design of caching and execution,

    C. Su, H. Zhang, H. Zhao, W. Shen, B. Ai, Y. Li, K. Bian, and B. Cui, “Caliex: A disk-based large-scale gnn training system with joint design of caching and execution,” inInternational Conference on Data Engineering (ICDE), 2025

  8. [16]

    Capsule: an out-of-core training mechanism for colossal gnns,

    Y. Xiang, Z. Ding, R. Guo, S. Wang, X. Xie, and S. K. Zhou, “Capsule: an out-of-core training mechanism for colossal gnns,” inProceedings of the ACM on Management of Data (SIGMOD), 2025

  9. [17]

    Outre: An out-of-core de-redundancy gnn training framework for massive graphs within a single machine,

    Z. Sheng, W. Zhang, Y. Tao, and B. Cui, “Outre: An out-of-core de-redundancy gnn training framework for massive graphs within a single machine,” inProceedings of the VLDB Endowment, 2024

  10. [18]

    Ginex: Ssd-enabled billion-scale graph neural network training on a single machine via provably optimal in-memory caching,

    Y. Park, S. Min, and J. W. Lee, “Ginex: Ssd-enabled billion-scale graph neural network training on a single machine via provably optimal in-memory caching,” inProceedings of the VLDB Endowment, 2022

  11. [19]

    Lumos: Dependency-driven disk-based graph processing,

    K. Vora, “Lumos: Dependency-driven disk-based graph processing,” inUSENIX Annual Technical Conference (USENIX ATC), 2019

  12. [20]

    X-stream: Edge-centric graph processing us- ing streaming partitions,

    A. Roy, I. Mihailovic, and W. Zwaenepoel, “X-stream: Edge-centric graph processing us- ing streaming partitions,” inACM Symposium on Operating Systems Principles (SOSP), 2013

  13. [21]

    Hy- perion: Co-optimizing ssd access and gpu computation for cost-efficient gnn training,

    J. Sun, M. Sun, Z. Zhang, Z. Shi, J. Xie, Z. Yang, J. Zhang, Z. Wang, and F. Wu, “Hy- perion: Co-optimizing ssd access and gpu computation for cost-efficient gnn training,” inIEEE International Conference on Data Engineering (ICDE), 2025

  14. [22]

    Accelerating sampling and aggregation operations in gnn frameworks with gpu initiated direct storage accesses,

    J. B. Park, V. S. Mailthody, Z. Qureshi, and W.-m. Hwu, “Accelerating sampling and aggregation operations in gnn frameworks with gpu initiated direct storage accesses,” in Proceedings of the VLDB Endowment, 2024

  15. [23]

    Open graph benchmark: Datasets for machine learning on graphs,

    W. Hu, M. Fey, M. Zitnik, Y. Dong, H. Ren, B. Liu, M. Catasta, and J. Leskovec, “Open graph benchmark: Datasets for machine learning on graphs,” inInternational Conference on Neural Information Processing Systems (NIPS), 2020

  16. [24]

    Inferturbo: A scalable system for boosting full-graph inference of graph neural network overhugegraphs,

    D. Zhang, X. Song, Z. Hu, Y. Li, M. Tao, B. Hu, L. Wang, Z. Zhang, and J. Zhou, “Inferturbo: A scalable system for boosting full-graph inference of graph neural network overhugegraphs,” inIEEE International Conference on Data Engineering (ICDE),2023

  17. [25]

    Accelerating training and inference of graph neural networks with fast sampling and pipelining,

    T. Kaler, N. Stathas, A. Ouyang, A.-S. Iliopoulos, T. Schardl, C. E. Leiserson, and J. Chen, “Accelerating training and inference of graph neural networks with fast sampling and pipelining,”Proceedings of Machine Learning and Systems (MLSys), 2022. 23

  18. [26]

    A gpu is all you need: Rethinking distributed and out-of- core gnn training,

    P. Naman and Y. Simmhan, “A gpu is all you need: Rethinking distributed and out-of- core gnn training,” inIEEE International Conference on High Performance Computing, Data and Analytics Workshop (HiPCW), 2025

  19. [27]

    Deep graph library: Towards efficient and scalable deep learning on graphs,

    M. Y. Wang, “Deep graph library: Towards efficient and scalable deep learning on graphs,” inICLR Workshop on Representation Learning on Graphs and Manifolds, 2019

  20. [28]

    Dgi: An easy and efficient framework for gnn model evaluation,

    P. Yin, X. Yan, J. Zhou, Q. Fu, Z. Cai, J. Cheng, B. Tang, and M. Wang, “Dgi: An easy and efficient framework for gnn model evaluation,” inACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023

  21. [29]

    How powerful are graph neural networks?

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful are graph neural networks?” inInternational Conference on Learning Representations (ICLR), 2019

  22. [30]

    Graph Attention Networks,

    P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y. Bengio, “Graph Attention Networks,” inInternational Conference on Learning Representations (ICLR), 2018

  23. [31]

    A linear time implementation of the reverse cuthill-mckee algorithm,

    W.-M. Chan and A. George, “A linear time implementation of the reverse cuthill-mckee algorithm,”BIT Numerical Mathematics, 1980

  24. [32]

    Stanford network analysis project,

    J. Leskovec, “Stanford network analysis project,” https://snap.stanford.edu/, 2020

  25. [33]

    Graph convolutional neural networks for web-scale recommender systems,

    R. Ying, R. He, K. Chen, P. Eksombatchai, W. L. Hamilton, and J. Leskovec, “Graph convolutional neural networks for web-scale recommender systems,” inACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2018

  26. [34]

    Temporal graph networks for deep learning on dynamic graphs,

    E. Rossi, B. Chamberlain, F. Frasca, D. Eynard, F. Monti, and M. Bronstein, “Temporal graph networks for deep learning on dynamic graphs,”arXiv preprint arXiv:2006.10637, 2020

  27. [35]

    Redundancy-free high- performance dynamic gnn training with hierarchical pipeline parallelism,

    Y. Xia, Z. Zhang, H. Wang, D. Yang, X. Zhou, and D. Cheng, “Redundancy-free high- performance dynamic gnn training with hierarchical pipeline parallelism,” inInterna- tional Symposium on High-Performance Parallel and Distributed Computing (HPDC), 2023

  28. [36]

    P3: Distributed deep graph learning at scale,

    S. Gandhi and A. P. Iyer, “P3: Distributed deep graph learning at scale,” inUSENIX Symposium on Operating Systems Design and Implementation (OSDI), 2021

  29. [37]

    Distdgl: Distributed graph neural network training for billion-scale graphs,

    D. Zheng, C. Ma, M. Wang, J. Zhou, Q. Su, X. Song, Q. Gan, Z. Zhang, and G. Karypis, “Distdgl: Distributed graph neural network training for billion-scale graphs,” inIEEE/ACM Workshop on Irregular Applications: Architectures and Algorithms (IA3), 2020

  30. [38]

    Fast graph representation learning with pytorch geometric,

    M. Fey and J. E. Lenssen, “Fast graph representation learning with pytorch geometric,” arXiv preprint arXiv:1903.02428, 2019

  31. [39]

    Mariusgnn: Resource- efficient out-of-core training of graph neural networks,

    R. Waleffe, J. Mohoney, T. Rekatsinas, and S. Venkataraman, “Mariusgnn: Resource- efficient out-of-core training of graph neural networks,” inEuropean Conference on Com- puter Systems (EuroSys), 2023. 24

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.