Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

SQUASH: Serverless and Distributed Quantization-based Attributed Vector Similarity Search

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a fully serverless quantization-based index can deliver hybrid vector search at 97 percent recall, up to 18x higher throughput, and up to 9x lower cost than commercial and server baselines.

desk verdict The OSQ segment-packing idea is a real contribution, but the paper's speedup and cost claims are unverifiable because it calibrates recall to 97% without ever reporting a measured recall. read the letter →

arxiv 2502.01528 v1 pith:MX3VQLJE submitted 2025-02-03 cs.DC cs.DB

Joe Oakley , Hakan Ferhatosmanoglu This is my paper · ORCID
classification cs.DCcs.DB
keywords vectorsimilaritysearchhybridattributefilteringscalarquantizationserverlesscomputingfunction-as-a-serviceapproximatenearestneighbordistributedindexing
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 vector similarity search with attribute filters—hybrid search—does not need dedicated servers or proximity graphs, and that a carefully packed scalar-quantization index can run it entirely on ephemeral cloud functions. It introduces SQUASH, which the authors describe as the first fully serverless distributed vector search system with native support for hybrid queries. The central promise is that the system reaches 97 percent recall on standard benchmarks while achieving up to 18x higher query throughput and up to 9x lower cost than a commercial serverless vector database and its own server-based baselines. If true, this would make elastic, pay-per-query hybrid search practical for workloads that scale to zero between bursts and for retrieval-augmented generation and recommendation tasks.

What carries the argument

The load-bearing object is OSQ, Optimized Scalar Quantization: non-uniform scalar quantization in which variable-length bit codes for consecutive dimensions are concatenated into shared S-bit segments, eliminating padding waste and enabling bit-shift extraction of any dimension. A low-bit variant assigns one bit per dimension, packs S dimensions into each S-bit segment, and uses Hamming distance for early pruning. Quantized attributes are handled the same way, with pass/fail bitmaps combined by bitwise AND, and a partition-vector map plus a centroid-distance threshold T selects partitions in one pass. Around this index, SQUASH adds a tree-based synchronous invocation scheme for launching thousands of function-as-a-service instances and a data-retention mechanism that reuses index data in warm containers.

What would settle it

Measure the rank correlation between one-bit OSQ Hamming distance and full Euclidean distance on a held-out set of query-vector pairs for a high-LID or attribute-correlated dataset; if the correlation is weak, or if recall at k = 10 falls below the 97 percent target when H_perc = 10, the early-pruning assumption fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a distributed scalar-quantization index can carry both vector distances and attribute filters, and that a one-bit-per-dimension version of that index yields Hamming distances whose ordering tracks the ordering of full Euclidean distances. That correlation lets SQUASH discard about 90 percent of candidate vectors with cheap bitwise comparisons before any floating-point distance is computed. Because the full-precision vectors never need to sit in memory, the index fits inside the memory and time limits of individual cloud-function instances, and the remaining candidates are ranked with quantized lower-bound distances plus a small re-ranking factor, R = 2. The result is a single-pass, partition-filtered search that the authors claim outperforms both a commercial serverless vector database and provisioned servers while meeting a 97 percent recall target.

Load-bearing premise

The load-bearing premise is that Hamming distance on the one-bit OSQ codes preserves the relative ordering of true Euclidean distances well enough that discarding the 90 percent of candidates with the largest Hamming distances does not drop true neighbors.

Editorial extensions

If this is right

  • If the central claim holds, hybrid vector search with arbitrary multi-attribute predicates can be offered as a serverless service that scales to zero and charges only for actual queries.
  • OSQ's low re-ranking requirement (R = 2) implies that memory-frugal quantization alone can sustain high recall, removing the need for proximity graphs or large full-precision caches in the search path.
  • The single-pass partition selection implies that filtered queries need only one distributed round trip, so latency stays predictable even as the number of partitions grows.
  • Data retention exploitation implies that repeated invocation of the same cloud-function container can skip most object-storage reads, cutting both latency and per-query cost on warm paths.
  • The cost model implies there is a crossover volume beyond which provisioned servers become cheaper than serverless; below roughly a million queries per day, serverless wins on cost.

Reading between the lines

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

  • The 97 percent recall figure is tied to a 10 percent Hamming-distance cutoff, H_perc = 10; on datasets with different intrinsic dimensionality or with attributes correlated with vector clusters, the cutoff would likely need retuning, and the paper's 'experiments not shown' would need to be made public to confirm the correlation.
  • The same segment-packing idea could be applied to other compressed nearest-neighbor settings, such as GPU batch search or embedding indexes inside database engines, where bit-shift extraction and SIMD lookups already pay off.
  • The tree-based invocation scheme is described for query workloads; a natural test would be to push it to update and insert workloads, where writes must also reach the right partitions without a central coordinator.
  • Because the cost model treats only query costs, the comparison to servers excludes index build and maintenance; a user should add build costs for a fair total-cost-of-ownership comparison.
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

4 major / 5 minor

Summary. The paper presents SQUASH, a serverless and distributed vector similarity search system implemented on AWS Lambda, targeting hybrid queries that combine vector similarity with attribute filters. The core technical contributions are OSQ, a segment-based scalar quantization scheme with non-uniform bit allocation and low-bit Hamming pruning; a single-pass filtered partition-ranking algorithm (Algorithm 1); a tree-based synchronous FaaS invocation scheme (Algorithm 2); and a data-retention mechanism for warm containers. The evaluation compares SQUASH with a commercial serverless vector database (System-X), the FaaS-based Vexless system, and EC2 server baselines, reporting up to 18x higher throughput and roughly 9x cost savings. The paper states that SQUASH is calibrated to achieve the same 97% recall as System-X, but it does not report any measured recall values.

Significance. If the central claims hold, the paper would make a useful systems contribution: OSQ's segment packing gives a concrete way to realize variable-length scalar quantization without per-dimension padding, the attribute-mask and partition-selection pipeline is a clean fit for FaaS parallelism, the tree-based invocation scheme addresses a real bottleneck in high-parallelism serverless communication, and the DRE mechanism for warm containers is practical. The paper also provides a cost model that is more detailed than typical for this area. The significance is however conditional on the accuracy evidence: the performance and cost comparisons are interpretable only if SQUASH actually achieves the recall level of the baselines, and the manuscript currently offers calibration statements rather than measurements. The algorithmic descriptions are generally clear and plausible, but the evaluation section is missing the one quantity that ties the whole comparison together.

major comments (4)
  1. [Section 5.1 and Section 5.3] The paper defines recall@k in Section 5.1 but never reports a single measured recall value for any dataset, configuration, or baseline. Section 5.3 states that 'SQUASH is calibrated to achieve the same 97% recall as System-X' and lists parameters (H_perc=10, R=2, T=1.13-1.2, beta=0.001), but no table or figure shows the achieved recall. This is load-bearing because the throughput and cost comparisons in Figures 8 and 9 are only apples-to-apples if accuracy parity is demonstrated. If the actual recall is below 97%, the reported QPS and cost advantages do not support the headline conclusion. The manuscript needs a recall@k table for every dataset and every comparison point, including the Vexless comparison in Table 3, with run-to-run variance.
  2. [Section 2.4.3] The fast-pruning mechanism relies on the assumption that Hamming distance on the one-bit OSQ representation preserves the relative ordering of Euclidean distances well enough that discarding 90% of candidates (H_perc=10) does not drop true neighbors. The text says 'Experiments (not shown) indicate strong correlations between the query-to-vector Euclidean distance and the Hamming distances computed based on binary OSQ vectors on a range of datasets.' This is an explicit admission that the key pruning criterion is not empirically verified within the paper. Since H_perc is a tuned parameter used to hit the 97% recall target, the paper should include either the correlation analysis or a recall-versus-H_perc ablation on the four evaluation datasets.
  3. [Section 5.2 and Section 5.3] The server-based baselines are not independent systems: Section 5.2 states that 'the same codebase as SQUASH is used, modified to run on a single machine (i.e., spawning separate processes rather than invoking parallel Lambda functions).' This means the comparison against 'Server-Small' and 'Server-Large' measures the effect of the FaaS scaling layer on the same implementation, not the competitiveness of SQUASH against established server-based hybrid ANNS systems such as FAISS IVFSQ, Milvus, or Filtered-DiskANN. The claims about 'significant performance improvements ... compared to server-based baselines' should be re-scoped, or the comparison should include independently implemented baselines with matched recall.
  4. [Section 5.3 and Equation (1)] The partition-selection threshold T in Equation (1) is calibrated per dataset (T=1.15, 1.2, 1.15, 1.13), with beta=0.001 fixed, and H_perc and R are chosen to meet the recall target. The paper gives no sensitivity analysis for T, beta, H_perc, or R, so it is possible that the 97% recall result is fitted to these datasets and will not transfer to other data distributions or attribute selectivities. The authors should report how recall, QPS, and cost vary with each of these parameters, and justify the chosen operating points rather than presenting them as defaults.
minor comments (5)
  1. [Section 5.3] The labels 'Server-S' and 'Server-L' in Figures 9 and 10 are not defined in the text; Section 5.3 only introduces c7i.4xlarge and c7i.16xlarge. Please use consistent labels or add a caption note.
  2. [Table 3] The table header 'recall=0.97' reads as a measured value, but the body text says 'We use the same recall target as in our other experiments.' Please mark it explicitly as a target or assumption, and report the measured recall for both SQUASH and Vexless.
  3. [Section 5.4] The cost comparison assumes that two server instances are provisioned for the server baselines to handle bursty traffic and redundancy. This assumption strongly affects the crossover points in Figure 8, but no sensitivity analysis or justification is given for the 'two instances' choice.
  4. [References] References [19] and [20] are duplicates (both cite RaBitQ with the same arXiv identifier), and references [74] and [75] are also duplicates. Please remove the repeated entries.
  5. [Section 3.5] The cost model in Equations (5) and (6) omits the cost of EFS Provisioned Throughput or burst credits, which can be significant if the optional fine-tuning stage performs many random reads. A sentence acknowledging this simplification would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the derivation is self-contained, though recall calibration and pruning evidence are under-reported.

full rationale

The paper's chain is algorithmic rather than derivational: OSQ bit allocation uses a variance-based iterative procedure, segment packing is a storage layout, attribute filtering is bitmask lookup, partition selection is a heuristic with a tunable threshold T and beta, low-bit pruning is a Hamming-distance heuristic, and final ranking uses lower-bound asymmetric distance with optional full-precision re-ranking. None of these steps takes the reported QPS or cost numbers as an input, and no equation reduces to the target recall. The only near-circular element is Section 5.3, where the recall level is a calibration target: "In all cases, SQUASH is calibrated to achieve the same 97% recall as System-X ... H_perc = 10, Fine-Tuning Ratio R = 2 ... beta = 0.001 ... T = 1.15 ...". This is a fitted parameter setting, not a predicted outcome; the paper does not disguise it as a measured result, although it also does not report any measured recall values. Similarly, Section 2.4.3 justifies the pruning correlation with "Experiments (not shown)". These are documentation and verification gaps, which create correctness risk (the speedup/cost comparison is only meaningful at equal recall), but they are not circular derivations because the recall figure is imposed as an input rather than derived from the system. Self-citations to prior quantization work and FSD-Inference are background or design inspiration and are not load-bearing for the central claim. The comparison against external datasets and baselines gives the performance/cost claims independent content.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central performance claim rests on several fitted thresholds and on an unshown correlation between binary Hamming distances and Euclidean ordering. The mathematically clean parts, namely bit packing, filtered-k traversal, and unitary transform preservation, do not by themselves deliver the reported recall.

free parameters (5)
  • Centroid distance threshold T = 1.15 (SIFT1M), 1.2 (GIST1M), 1.15 (SIFT10M), 1.13 (DEEP10M)
    Used in Algorithm 1 to decide which partitions to visit; chosen per dataset to reach the 97 percent recall target in Section 5.3.
  • beta in the T formula = 0.001
    Tunable additive term in Equation (1), described as adjustable based on recall requirements in Section 2.4.2.
  • Hamming cutoff percentage H_perc = 10
    Fraction of local candidates retained after low-bit OSQ Hamming pruning in Section 2.4.3; set to 10 for all datasets to achieve 97 percent recall.
  • Fine-tuning ratio R = 2
    Multiplier for full-precision records fetched in post-refinement in Section 2.4.5; R=2 is used in all experiments.
  • Quantization bit budget b = 4 x d bits per vector
    Total quantization budget per vector, chosen uniformly as four times dimensionality; it defines the compression accuracy tradeoff and is not derived from an accuracy target.
assumptions (3)
  • domain assumption Hamming distance on binary-quantized vectors approximately preserves Euclidean distance ordering.
    Stated in Section 2.4.3 as an observed correlation, with "Experiments (not shown)"; the 10 percent Hamming pruning step relies on this correlation, but no proof or visible experiment is provided.
  • standard math Unitary transforms such as KLT are distance-preserving, so combining results from independently transformed partitions gives correct query answers.
    Used in Section 2.4.1; this is mathematically correct for orthonormal transforms.
  • ad hoc to paper The partition selection threshold formula T = 1 + sigma_mu / mu_mu + beta / sqrt(d) yields high recall when paired with the stopping rule.
    Equation (1) is a heuristic construction based on distance-ratio statistics rather than a derived accuracy guarantee; its validity is established only by calibration to 97 percent recall in Section 5.3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SQUASH: Serverless and Distributed Quantization-based Attributed Vector Similarity Search." pith.science (2026). https://pith.science/paper/MX3VQLJE

@misc{pith2026250201528,
  author       = {Pith},
  title        = {Pith review of: SQUASH: Serverless and Distributed Quantization-based Attributed Vector Similarity Search},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MX3VQLJE}},
  note         = {Machine review of arXiv:2502.01528}
}
read the original abstract

Vector similarity search presents significant challenges in terms of scalability for large and high-dimensional datasets, as well as in providing native support for hybrid queries. Serverless computing and cloud functions offer attractive benefits such as elasticity and cost-effectiveness, but are difficult to apply to data-intensive workloads. Jointly addressing these two main challenges, we present SQUASH, the first fully serverless vector search solution with rich support for hybrid queries. It features OSQ, an optimized and highly parallelizable quantization-based approach for vectors and attributes. Its segment-based storage mechanism enables significant compression in resource-constrained settings and offers efficient dimensional extraction operations. SQUASH performs a single distributed pass to guarantee the return of sufficiently many vectors satisfying the filter predicate, achieving high accuracy and avoiding redundant computation for vectors which fail the predicate. A multi-level search workflow is introduced to prune most vectors early to minimize the load on Function-as-a-Service (FaaS) instances. SQUASH is designed to identify and utilize retention of relevant data in re-used runtime containers, which eliminates redundant I/O and reduces costs. Finally, we demonstrate a new tree-based method for rapid FaaS invocation, enabling the bi-directional flow of data via request/response payloads. Experiments comparing SQUASH with state-of-the-art serverless vector search solutions and server-based baselines on vector search benchmarks confirm significant performance improvements at a lower cost.

Figures

Figures reproduced from arXiv: 2502.01528 by the authors.

Figure 1
Figure 1. Comparison of SQ and OSQ storage schemes [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Bit savings under OSQ vs SQ 2.2.2 OSQ Dimensional Extraction. In order to access individual dimensions from our OSQ index, we require a mechanism to effi￾ciently extract sub-𝑆-bit chunks from an 𝑆-bit segment. This func￾tionality is required when we seek to compute dimension-wise distances between a query vector and a set of data vectors. The OSQ extraction scheme utilizes column-wise lightweight left/right bit-shif… view at source ↗
Figure 3
Figure 3. Illustrative Example of OSQ Dimensional Extraction Procedure, with [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: SQUASH attribute filtering workflow balanced partitions for computational load balance in the resource￾constrained FaaS environment. Alternative balanced partitioning schemes can be utilized, for example fusing vector and multi-attribute similarity information. Within …
Figure 5
Figure 5. Figure 5: SQUASH high-level architecture squared ‘query dimension’ to ‘dimension boundary value’ distance to be calculated only once. Building 𝐿 requires only ( Í 𝑗 𝐶[𝑗]) − 1 calculations, and is performed extremely efficiently with vectorized operations. As described above, 𝐿[𝑗…
Figure 6
Figure 6. Figure 6: Cost, latency and S3 request reduction with DRE. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Tree-based FaaS invocation scheme. Blue circles [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Daily cost of SQUASH, System-X and small/large [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Queries per second (QPS) for SQUASH, System-X [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. QASP: Query-Adaptive Robust Vector Search Policy

    cs.IR 2026-07 reject novelty 6.0 of 10

    QASP predicts each query's full recall-vs-probes curve with one up-front regression and derives a per-query search depth for any recall target, cutting reported data access by up to 80% at 99% recall.

Reference graph

Works this paper leans on

82 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [1]

    Cecilia Aguerrebere, Ishwar Singh Bhati, Mark Hildebrand, Mariano Tepper, and Theodore Willke. 2023. Similarity Search in the Blink of an Eye with Compressed Indices. Proc. VLDB Endow. 16, 11 (jul 2023), 3433–3446. https: //doi.org/10.14778/3611479.3611537

    arXiv 2023

  2. [2]

    Cecilia Aguerrebere, Mark Hildebrand, Ishwar Singh Bhati, Theodore Willke, and Mariano Tepper. 2024. Locally-Adaptive Quantization for Streaming Vector Search. arXiv:2402.02044 [cs.LG] https://arxiv.org/abs/2402.02044

    arXiv 2024

  3. [3]

    Amazon. 2024. A WS Lambda Memory and Computing Power. https://docs.aws. amazon.com/lambda/latest/operatorguide/computing-power.html

    2024

  4. [4]

    Amazon. 2024. What’s New in Serverless, A WS re:Invent 2020 . https://www. youtube.com/watch?v=aW5EtKHTMuQ&t=339s

    2024

  5. [5]

    Amazon. 2025. A WS Lambda. https://aws.amazon.com/lambda/

    work page 2025

  6. [6]

    Alexandr Andoni and Ilya Razenshteyn. 2015. Optimal Data-Dependent Hashing for Approximate Near Neighbors. In Proceedings of the Forty-Seventh Annual ACM Symposium on Theory of Computing (Portland, Oregon, USA) (STOC ’15). Association for Computing Machinery, New York, NY, USA, 793–801. https: //doi.org/10.1145/2746539.2746553

    arXiv 2015

  7. [7]

    Fabien André, Anne-Marie Kermarrec, and Nicolas Le Scouarnec. 2015. Cache locality is not enough: high-performance nearest neighbor search with product quantization fast scan. Proc. VLDB Endow. 9, 4 (Dec. 2015), 288–299. https: //doi.org/10.14778/2856318.2856324

    arXiv 2015

  8. [8]

    Shivam Bhatele. 2023. Vectorization in Python - An Alternative to Python Loops. https://medium.com/pythoneers/vectorization-in-python-an-alternative- to-python-loops-2728d6d7cd3e

    work page 2023

Show all 82 references
  1. [9]

    James Briggs. 2025. Faiss: The Missing Manual . https://www.pinecone.io/learn/ series/faiss/

    2025

  2. [10]

    Datastax. 2024. Astra DB Serverless . https://docs.datastax.com/en/astra-db- serverless/index.html

    2024

  3. [11]

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. 2024. The Faiss library. arXiv:2401.08281 [cs.LG] https://arxiv.org/abs/2401.08281

    2024 arXiv

  4. [12]

    Karima Echihabi, Kostas Zoumpatianos, Themis Palpanas, and Houda Benbrahim

  5. [15]

    Hakan Ferhatosmanoglu, Ertem Tuncel, Divyakant Agrawal, and Amr El Abbadi

  6. [16]

    Cong Fu, Changxu Wang, and Deng Cai. 2021. High dimensional similarity search with satellite system graph: Efficiency, scalability, and unindexed query compatibility. IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 8 (2021), 4139–4150

    2021

  7. [17]

    Cong Fu, Chao Xiang, Changxu Wang, and Deng Cai. 2019. Fast approximate nearest neighbor search with the navigating spreading-out graph. Proc. VLDB Endow. 12, 5 (Jan. 2019), 461–474. https://doi.org/10.14778/3303753.3303754

    2019

  8. [18]

    Jianyang Gao and Cheng Long. 2023. High-Dimensional Approximate Nearest Neighbor Search: with Reliable and Efficient Distance Comparison Operations. Proc. ACM Manag. Data 1, 2, Article 137 (jun 2023), 27 pages. https://doi.org/10. 1145/3589282

    2023

  9. [20]

    Jianyang Gao and Cheng Long. 2024. RaBitQ: Quantizing High-Dimensional Vectors with a Theoretical Error Bound for Approximate Nearest Neighbor Search. Proc. ACM Manag. Data 2, 3, Article 167 (May 2024), 27 pages. https: //doi.org/10.1145/3654970

    2024 doi

  10. [21]

    Tiezheng Ge, Kaiming He, Qifa Ke, and Jian Sun. 2014. Optimized Product Quantization. IEEE Transactions on Pattern Analysis and Machine Intelligence 36, 4 (2014), 744–755. https://doi.org/10.1109/TPAMI.2013.240

    2014 doi

  11. [22]

    2012.Vector quantization and signal compression

    Allen Gersho and Robert M Gray. 2012.Vector quantization and signal compression. Vol. 159. Springer Science & Business Media

    2012

  12. [23]

    Siddharth Gollapudi, Neel Karia, Varun Sivashankar, Ravishankar Krishnaswamy, Nikit Begwani, Swapnil Raz, Yiyong Lin, Yin Zhang, Neelam Mahapatro, Premku- mar Srinivasan, Amit Singh, and Harsha Vardhan Simhadri. 2023. Filtered- DiskANN: Graph Algorithms for Approximate Nearest...

    2023

  13. [24]

    Gaurav Gupta, Jonah Yi, Benjamin Coleman, Chen Luo, Vihan Lakshman, and Anshumali Shrivastava. 2023. CAPS: A Practical Partition Index for Filtered Similarity Search. arXiv:2308.15014 [cs.IR] https://arxiv.org/abs/2308.15014

    2023 arXiv

  14. [25]

    Houle and Michael Nett

    Michael E. Houle and Michael Nett. 2015. Rank-Based Similarity Search: Reducing the Dimensional Dependence. IEEE Transactions on Pattern Analysis and Machine Intelligence 37, 1 (2015), 136–150. https://doi.org/10.1109/TPAMI.2014.2343223

    2015

  15. [26]

    Piotr Indyk and Rajeev Motwani. 1998. Approximate nearest neighbors: towards removing the curse of dimensionality. InProceedings of the Thirtieth Annual ACM Symposium on Theory of Computing (Dallas, Texas, USA) (STOC ’98). Association for Computing Machinery, New York, NY, USA...

    1998

  16. [27]

    Shikhar Jaiswal, Ravishankar Krishnaswamy, Ankit Garg, Harsha Vardhan Simhadri, and Sheshansh Agrawal. 2022. OOD-DiskANN: Efficient and Scal- able Graph ANNS for Out-of-Distribution Queries. arXiv:2211.12850 [cs.LG] https://arxiv.org/abs/2211.12850

    2022 arXiv

  17. [28]

    Jananie Jarachanthan, Li Chen, Fei Xu, and Bo Li. 2021. AMPS-Inf: Automatic Model Partitioning for Serverless Inference with Cost Efficiency. In Proceedings of the 50th International Conference on Parallel Processing (Lemont, IL, USA) (ICPP ’21). Association for Computing Mach...

    2021

  18. [29]

    Suhas Jayaram Subramanya, Fnu Devvrit, Harsha Vardhan Simhadri, Ravishankar Krishnawamy, and Rohan Kadekodi. 2019. DiskANN: Fast Accurate Billion-point Nearest Neighbor Search on a Single Node. InAdvances in Neural Information Pro- cessing Systems, H. Wallach, H. Larochelle, A...

    2019

  19. [30]

    Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2021. Billion-Scale Similarity Search with GPUs. IEEE Transactions on Big Data 7, 3 (2021), 535–547. https: //doi.org/10.1109/TBDATA.2019.2921572

    2021

  20. [31]

    Herve Jégou, Matthijs Douze, and Cordelia Schmid. 2011. Product Quantization for Nearest Neighbor Search. IEEE Transactions on Pattern Analysis and Machine Intelligence 33, 1 (2011), 117–128. https://doi.org/10.1109/TPAMI.2010.57

    2011 doi

  21. [32]

    Zijun Li, Linsong Guo, Jiagan Cheng, Quan Chen, Bingsheng He, and Minyi Guo. 2022. The Serverless Computing Survey: A Technical Primer for Design Architecture. ACM Computing Survey 54, 10s, Article 220 (sep 2022), 34 pages. https://doi.org/10.1145/3508360

    2022 doi

  22. [33]

    Stuart Lloyd. 1982. Least squares quantization in PCM. IEEE transactions on information theory 28, 2 (1982), 129–137

    1982

  23. [34]

    Kejing Lu, Hongya Wang, Wei Wang, and Mineichi Kudo. 2020. VHP: approxi- mate nearest neighbor search via virtual hypersphere partitioning. Proc. VLDB Endow. 13, 9 (May 2020), 1443–1455. https://doi.org/10.14778/3397230.3397240

    2020

  24. [35]

    Qin Lv, William Josephson, Zhe Wang, Moses Charikar, and Kai Li. 2017. Intelli- gent probing for locality sensitive hashing: multi-probe LSH and beyond. Proc. VLDB Endow. 10, 12 (Aug. 2017), 2021–2024. https://doi.org/10.14778/3137765. 3137836

    2017 doi

  25. [36]

    Yury Malkov, Alexander Ponomarenko, Andrey Logvinov, and Vladimir Krylov

  26. [37]

    Malkov and D

    Yu A. Malkov and D. A. Yashunin. 2020. Efficient and Robust Approximate Nearest Neighbor Search Using Hierarchical Navigable Small World Graphs. IEEE Transactions on Pattern Analysis and Machine Intelligence 42, 4 (2020), 824–

    2020

  27. [38]

    Eric Martin and Eddie Cao. 2015. Euclidean chemical spaces from molecular fingerprints: Hamming distance and Hempel’s ravens.Journal of Computer-Aided Molecular Design 29, 5 (01 May 2015), 387–395. https://doi.org/10.1007/s10822- 014-9819-y

    2015 doi

  28. [39]

    Sanparith Marukatat and Ithipan Methasate. 2013. Fast nearest neighbor retrieval using randomized binary codes and approximate Euclidean distance. Pattern Recognition Letters 34, 9 (2013), 1101–1107. https://doi.org/10.1016/j.patrec.2013. 03.006

    2013 doi

  29. [40]

    Microsoft. [n.d.]. Azure Functions Overview. https://azure.microsoft.com/en- gb/products/functions Accessed: 2023-11-30

    2023

  30. [41]

    Marius Muja and David G. Lowe. 2014. Scalable Nearest Neighbor Algorithms for High Dimensional Data. IEEE Transactions on Pattern Analysis and Machine In- telligence 36, 11 (2014), 2227–2240. https://doi.org/10.1109/TPAMI.2014.2321376

    2014

  31. [42]

    Ingo Müller, Renato Marroquín, and Gustavo Alonso. 2020. Lambada: Interactive Data Analytics on Cold Data Using Serverless Cloud Infrastructure. In Proceed- ings of the 2020 ACM SIGMOD International Conference on Management of Data (Portland, OR, USA) (SIGMOD ’20). Association...

    2020

  32. [43]

    Lushuai Niu, Zhi Xu, Longyang Zhao, Daojing He, Jianqiu Ji, Xiaoli Yuan, and Mian Xue. 2023. Residual Vector Product Quantization for approximate nearest neighbor search. Expert Systems with Applications 232 (2023), 120832. https: //doi.org/10.1016/j.eswa.2023.120832 Joe Oakle...

    2023

  33. [44]

    Haechan Noh, Taeho Kim, and Jae-Pil Heo. 2021. Product Quantizer Aware In- verted Index for Scalable Nearest Neighbor Search. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) . 12210–12218

    2021

  34. [45]

    NumPy. 2025. Indexing on ndarrays. https://numpy.org/doc/stable/user/basics. indexing.html

    2025

  35. [46]

    Joe Oakley, Chris Conlan, Gunduz Vehbi Demirci, Alexandros Sfyridis, and Hakan Ferhatosmanoglu. 2024. Foresight plus: serverless spatio-temporal traffic forecasting. GeoInformatica (26 Apr 2024). https://doi.org/10.1007/s10707-024- 00517-9

    2024 doi

  36. [47]

    Joe Oakley and Hakan Ferhatosmanoglu. 2024. FSD-Inference: Fully Serverless Distributed Inference with Scalable Cloud Communication. In 2024 IEEE 40th International Conference on Data Engineering (ICDE) . 2109–2122. https://doi.org/ 10.1109/ICDE60146.2024.00168

    2024

  37. [48]

    Elmore, and Michael J

    John Paparrizos, Ikraduya Edian, Chunwei Liu, Aaron J. Elmore, and Michael J. Franklin. 2022. Fast Adaptive Similarity Search through Variance-Aware Quanti- zation. In 2022 IEEE 38th International Conference on Data Engineering (ICDE) . 2969–2983. https://doi.org/10.1109/ICDE5...

    2022

  38. [49]

    Yongjoo Park, Michael Cafarella, and Barzan Mozafari. 2015. Neighbor-sensitive hashing. Proc. VLDB Endow. 9, 3 (Nov. 2015), 144–155. https://doi.org/10.14778/ 2850583.2850589

    2015

  39. [50]

    Liana Patel, Peter Kraft, Carlos Guestrin, and Matei Zaharia. 2024. ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Struc- tured Data. Proc. ACM Manag. Data 2, 3, Article 120 (may 2024), 27 pages. https://doi.org/10.1145/3654923

    2024 doi

  40. [51]

    Matthew Perron, Raul Castro Fernandez, David DeWitt, and Samuel Madden

  41. [52]

    Pinecone. 2024. Pinecone Serverless. https://www.pinecone.io/product/

    2024

  42. [53]

    Moneer Rifai. [n.d.]. Serverless showdown: A WS Lambda vs Azure Functions vs Google Cloud Functions . https://www.pluralsight.com/resources/blog/cloud/ serverless-showdown-aws-lambda-vs-azure-functions-vs-google-cloud- functions Accessed: 2023-11-30

    2023

  43. [54]

    Chanop Silpa-Anan and Richard Hartley. 2008. Optimised KD-trees for fast image descriptor matching. In 2008 IEEE Conference on Computer Vision and Pattern Recognition. 1–8. https://doi.org/10.1109/CVPR.2008.4587638

    2008

  44. [55]

    Aditi Singh, Suhas Jayaram Subramanya, Ravishankar Krishnaswamy, and Har- sha Vardhan Simhadri. 2021. FreshDiskANN: A Fast and Accurate Graph- Based ANN Index for Streaming Similarity Search. arXiv:2105.09613 [cs.IR] https://arxiv.org/abs/2105.09613

    2021 arXiv

  45. [56]

    Yongye Su, Yinqi Sun, Minjia Zhang, and Jianguo Wang. 2024. Vexless: A Server- less Vector Data Management System Using Cloud Functions. Proc. ACM Manag. Data 2, 3, Article 187 (may 2024), 26 pages. https://doi.org/10.1145/3654990

    2024 doi

  46. [57]

    Franklin, Sanjay Krishnan, and Reynold S

    Liwen Sun, Michael J. Franklin, Sanjay Krishnan, and Reynold S. Xin. 2014. Fine-grained partitioning for aggressive data skipping. In Proceedings of the 2014 ACM SIGMOD International Conference on Management of Data (Snowbird, Utah, USA) (SIGMOD ’14). Association for Computing...

    2014

  47. [58]

    Yao Tian, Xi Zhao, and Xiaofang Zhou. 2023. DB-LSH 2.0: Locality-sensitive hashing with query-based dynamic bucketing. IEEE Transactions on Knowledge and Data Engineering (2023)

    2023

  48. [59]

    Ertem Tuncel, Hakan Ferhatosmanoglu, and Kenneth Rose. 2002. VQ-index: an index structure for similarity searching in multimedia databases. In Proceedings of the Tenth ACM International Conference on Multimedia (Juan-les-Pins, France) (MULTIMEDIA ’02). Association for Computin...

    2002

  49. [60]

    TurboPuffer. 2024. TurboPuffer. https://turbopuffer.com

    2024

  50. [61]

    Upstash. 2024. Upstash Vector: Serverless Vector Database for AI and LLMs . https: //upstash.com/blog/introducing-vector-database

    2024

  51. [62]

    Jianguo Wang, Xiaomeng Yi, Rentong Guo, Hai Jin, Peng Xu, Shengjun Li, Xiangyu Wang, Xiangzhou Guo, Chengming Li, Xiaohai Xu, Kun Yu, Yuxing Yuan, Yinghao Zou, Jiquan Long, Yudong Cai, Zhenxiang Li, Zhifeng Zhang, Yihua Mo, Jun Gu, Ruiyi Jiang, Yi Wei, and Charles Xie. 2021. M...

    2021

  52. [63]

    Mengzhao Wang, Lingwei Lv, Xiaoliang Xu, Yuxiang Wang, Qiang Yue, and Jiongkang Ni. 2022. Navigable Proximity Graph-Driven Native Hybrid Queries with Structured and Unstructured Constraints. arXiv:2203.13601 [cs.DB] https: //arxiv.org/abs/2203.13601

    2022 arXiv

  53. [64]

    Mengzhao Wang, Lingwei Lv, Xiaoliang Xu, Yuxiang Wang, Qiang Yue, and Jiongkang Ni. 2023. An Efficient and Robust Framework for Ap- proximate Nearest Neighbor Search with Attribute Constraint. In Advances in Neural Information Processing Systems , A. Oh, T. Naumann, A. Glober-...

    2023

  54. [65]

    Mengzhao Wang, Weizhi Xu, Xiaomeng Yi, Songlin Wu, Zhangyang Peng, Xi- angyu Ke, Yunjun Gao, Xiaoliang Xu, Rentong Guo, and Charles Xie. 2024. Starling: An I/O-Efficient Disk-Resident Graph Index Framework for High- Dimensional Vector Similarity Search on Data Segment. Proc. A...

    2024 doi

  55. [66]

    Runhui Wang and Dong Deng. 2020. DeltaPQ: lossless product quantization code compression for high dimensional similarity search. Proc. VLDB Endow. 13, 13 (sep 2020), 3603–3616. https://doi.org/10.14778/3424573.3424580

    2020

  56. [67]

    Weaviate. 2024. Weaviate Serverless Database. https://weaviate.io/deployment/ serverless

    2024

  57. [68]

    Roger Weber, Hans-Jörg Schek, and Stephen Blott. 1998. A Quantitative Analysis and Performance Study for Similarity-Search Methods in High-Dimensional Spaces. In Proceedings of the 24rd International Conference on Very Large Data Bases (VLDB ’98). Morgan Kaufmann Publishers In...

    1998

  58. [69]

    Chuangxian Wei, Bin Wu, Sheng Wang, Renjie Lou, Chaoqun Zhan, Feifei Li, and Yuanzhe Cai. 2020. AnalyticDB-V: a hybrid analytical engine towards query fusion for structured and unstructured data. Proc. VLDB Endow. 13, 12 (aug 2020), 3152–3165. https://doi.org/10.14778/3415478.3415541

    2020

  59. [70]

    Xiaoliang Xu, Chang Li, Yuxiang Wang, and Yixing Xia. 2020. Mul- tiattribute approximate nearest neighbor search based on naviga- ble small world graph. Concurrency and Computation: Practice and Experience 32, 24 (2020), e5970. https://doi.org/10.1002/cpe.5970 arXiv:https://on...

    2020 doi

  60. [71]

    Minchen Yu, Zhifeng Jiang, Hok Chun Ng, Wei Wang, Ruichuan Chen, and Bo Li. 2021. Gillis: Serving Large Neural Networks in Serverless Functions with Automatic Model Partitioning. In 2021 IEEE 41st International Conference on Distributed Computing Systems (ICDCS) . 138–148. htt...

    2021

  61. [72]

    Qianxi Zhang, Shuotao Xu, Qi Chen, Guoxin Sui, Jiadong Xie, Zhizhen Cai, Yaoqi Chen, Yinxuan He, Yuqing Yang, Fan Yang, Mao Yang, and Lidong Zhou

  62. [73]

    Weijie Zhao, Shulong Tan, and Ping Li. 2020. SONG: Approximate Nearest Neighbor Search on GPU. In 2020 IEEE 36th International Conference on Data Engineering (ICDE). 1033–1044. https://doi.org/10.1109/ICDE48307.2020.00094

    2020

  63. [75]

    Weijie Zhao, Shulong Tan, and Ping Li. 2022. Constrained Approximate Similarity Search on Proximity Graph. arXiv:2210.14958 [cs.IR] https://arxiv.org/abs/2210. 14958

    2022 arXiv

  64. [76]

    Bolong Zheng, Xi Zhao, Lianggui Weng, Nguyen Quoc Viet Hung, Hang Liu, and Christian S. Jensen. 2020. PM-LSH: A fast and accurate LSH framework for high-dimensional approximate NN search. Proc. VLDB Endow. 13, 5 (Jan. 2020), 643–655. https://doi.org/10.14778/3377369.3377374

    2020

  65. [77]

    Zilliz. 2023. Scalar Quantization and Product Quantization . https://zilliz.com/ learn/scalar-quantization-and-product-quantization

    2023

  66. [78]

    Chaoji Zuo, Miao Qiao, Wenchao Zhou, Feifei Li, and Dong Deng. 2024. SeRF: Segment Graph for Range-Filtering Approximate Nearest Neighbor Search. Proc. ACM Manag. Data 2, 1, Article 69 (March 2024), 26 pages. https://doi.org/10. 1145/3639324

    2024

  67. [836]

    https://doi.org/10.1109/TPAMI.2018.2889473

    2018

  68. [2000]

    In Proceedings of the Ninth International Conference on Information and Knowledge Management (McLean, Virginia, USA) (CIKM ’00)

    Vector approximation based indexing for non-uniform high dimensional data sets. In Proceedings of the Ninth International Conference on Information and Knowledge Management (McLean, Virginia, USA) (CIKM ’00). Association for Computing Machinery, New York, NY, USA, 202–209. htt...

  69. [2001]

    In Proceedings 17th International Conference on Data Engineering

    Approximate nearest neighbor searching in multimedia databases. In Proceedings 17th International Conference on Data Engineering . IEEE, 503–511

  70. [2006]

    Information Systems 31, 6 (2006), 512–540

    High dimensional nearest neighbor searching. Information Systems 31, 6 (2006), 512–540. https://doi.org/10.1016/j.is.2005.01.001

    2006 doi

  71. [2014]

    Information Systems 45 (2014), 61–68

    Approximate nearest neighbor algorithm based on navigable small world graphs. Information Systems 45 (2014), 61–68. https://doi.org/10.1016/j.is.2013. 10.006

    2014 doi

  72. [2018]

    The lernaean hydra of data series similarity search: an experimental evaluation of the state of the art. Proc. VLDB Endow. 12, 2 (oct 2018), 112–127. https://doi.org/10.14778/3282495.3282498

    2018

  73. [2020]

    InProceedings of the 2020 ACM SIGMOD International Conference on Management of Data (Portland, OR, USA) (SIGMOD ’20)

    Starling: A Scalable Query Engine on Cloud Functions. InProceedings of the 2020 ACM SIGMOD International Conference on Management of Data (Portland, OR, USA) (SIGMOD ’20). Association for Computing Machinery, New York, NY, USA, 131–141. https://doi.org/10.1145/3318464.3380609

    2020

  74. [2023]

    In 17th USENIX Symposium on Operating Systems Design and Implementation (OSDI 23)

    VBASE: Unifying Online Vector Similarity Search and Relational Queries via Relaxed Monotonicity. In 17th USENIX Symposium on Operating Systems Design and Implementation (OSDI 23). USENIX Association, Boston, MA, 377–395. https://www.usenix.org/conference/osdi23/presentation/zh...

Pith tools

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