Pith. sign in

REVIEW 5 major objections 5 minor 55 references

ASAP: Reimagining the Data Lifecycle using Application Semantic-Aware Processing

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Sketching, sampling, rollups, and compression are one family of semantic-preserving primitives; treating them uniformly across a data pipeline can cut cost-scale-performance tradeoffs by up to 1000x.

desk verdict A clear, honest position paper whose unifying-frame claim is sound but whose headline numbers rest on extrapolation and synthetic remapping; worth engaging, but the quantitative evidence needs direct measurement. read the letter →

arxiv 2608.12735 v1 pith:3IWT3IEO submitted 2026-08-13 cs.DB cs.NIcs.PF

Milind Srivastava , Zeying Zhu , Yajie Zhou , Yancheng Yuan , Fenghao Dong , Peilin Xin , Zaoxing Liu , Vyas Sekar This is my paper · ORCID
classification cs.DBcs.NIcs.PF
keywords applicationsemantic-awareprocessingdatalifecycleCost-Scale-Performancetradeoffsketchingapproximatequerycompressionrollupspipelinesynthesis
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 identifies a common structure beneath apparently unrelated data-reduction techniques: each one preserves only the information an application needs, rather than the raw data itself. It proposes ASAP, a paradigm in which such semantic-preserving primitives become first-class citizens across the entire Collect-Transmit-Store-Analyze lifecycle, not isolated fixes for a single stage. The payoff, if the vision holds, is compounding: analytics run directly on sketches computed at the source, primitives invented for networking transfer to observability, and cost, scale, and performance improve together. Preliminary pipelines in metrics observability, distributed tracing, and network monitoring report up to three orders of magnitude improvement in the cost-scale-performance tradeoff, for example 48x lower transmission cost and 81x lower analytics cost in the metrics case.

What carries the argument

The load-bearing object is the ASAP primitive: an operation that maps raw data to a compact form while preserving the semantics an application needs—sampling preserves general summaries, sketches preserve aggregates, histograms preserve distributions, wavelets preserve frequency structure, rollups preserve hierarchies, and generative models capture high-dimensional relationships. The argument treats the Collect-Transmit-Store-Analyze lifecycle as the substrate on which primitives are placed, and the framework's data plane and control plane as the machinery that computes, stores, and reconfigures primitive instances. What carries the argument is the claim that primitives defined by the semantics they preserve can be composed across stages, for instance by building a sketch at the collector, transmitting the compact sketch, and serving dashboard queries from it. The prototypes instantiate this composition for metrics observability, distributed tracing, and network monitoring.

What would settle it

Run the ASAP metrics pipeline on a workload that includes joins, nested queries, or point lookups rather than only quantile, top-k, and cardinality aggregates; if accuracy falls below the stated 95 percent relative target or the claimed 48x and 81x cost reductions disappear, the cross-lifecycle benefit is confined to aggregate workloads. A second check is to compose two primitives, such as a wavelet transform followed by a sketch, and measure whether the downstream relative error exceeds the product of the individual error guarantees.

Watch

Extended reading notes

Core claim

The central claim is that the Cost-Scale-Performance (CSP) problem—low cost at large scale with high performance—is best attacked not by optimizing each lifecycle stage in isolation, but by viewing the data pipeline as an information-preservation problem. Every established technique, from sketches in network routers to AQP sampling in databases to rollups in time-series stores, is an instance of the same principle: process raw data into a form that keeps the properties an application cares about and discards the rest. The paper calls these ASAP primitives, groups them by the semantics they preserve rather than the domain they come from, and argues that a unified abstraction enables three new moves: reusing a primitive across domains, swapping primitives as workloads change, and stacking primitives across lifecycle stages so that benefits multiply. The paper's evidence is three implemented pipelines that combine primitives across stages; on these, the paper reports query speedups up to 3000x, memory reductions up to 7000x, transmission cost reductions of 48x, and storage reductions of 600x, all while meeting stated fidelity targets.

Load-bearing premise

The approach hinges on the assumption that a summary or sketch made at one stage still preserves everything a downstream stage needs; if composing such summaries across stages is lossy in general, the cross-lifecycle savings shrink, and the paper currently offers no guarantee to rule that out.

Editorial extensions

If this is right

  • Analytics can run directly on compact primitives such as sketches computed at the data source, removing the need to transmit or store raw data before answering aggregate queries.
  • A primitive proven in one domain can be reused with its fidelity guarantees and configuration logic in another domain, such as carrying a sketch designed for network monitoring into observability.
  • Pipelines become replaceable-component systems: as workloads or primitives evolve, the same pipeline can swap a sketch for a wavelet representation without rebuilding the surrounding stages.
  • If cross-stage composition preserves the information downstream queries need, CSP benefits compound across the lifecycle, which is how the paper obtains its up-to-three-orders-of-magnitude figures.
  • The framework's control plane can reconfigure primitives at runtime as data and query workloads shift, rather than fixing one configuration at startup.

Reading between the lines

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

  • The paper leaves implicit the strongest test of its thesis: whether composing semantic-preserving primitives across stages preserves downstream fidelity; a formal or empirical composition guarantee would turn the preliminary results into a general design law.
  • The paper leaves implicit the energy version of its argument: preserving only what applications need reduces computation, data movement, and storage, so a cost-aware synthesis objective could naturally minimize carbon footprint as well.
  • The paper leaves implicit that agentic data consumers, which tolerate approximation and demand low latency, are a natural early adopter class, since they can act directly on the error bounds and provenance of primitive instances.
  • A testable extension beyond the paper is to benchmark whether the three-orders-of-magnitude benefits persist on workloads beyond aggregates, such as joins or nested queries, where multiple primitives must compose within a single query plan.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper argues that diverse cost-reduction techniques (sketching, sampling, rollups, compression, wavelets) are all instances of "application semantic-aware processing" (ASAP), and proposes a paradigm in which these primitives are treated as first-class and unified across the Collect-Transmit-Store-Analyze lifecycle and across application domains. It motivates the paradigm through a survey of isolated per-stage solutions, presents an envisioned framework with problem/deployment specifications and automatic pipeline synthesis, and reports three prototype pipelines (metrics observability, distributed tracing, network monitoring) that show substantial cost/performance benefits over exact baselines such as Prometheus and ClickHouse. It closes with open challenges including primitive selection, cross-lifecycle placement, interoperability, reconfiguration, complex queries, and incremental adoption.

Significance. If the vision is realized, the paper identifies a genuinely useful unifying abstraction: it connects disparate communities under a common design principle and articulates concrete cross-lifecycle optimizations (e.g., answering queries on sketches built at the source) that individual per-stage efforts miss. The three prototypes and their comparisons to external baselines are a good start, and the paper is honest that the evidence is preliminary and that key composition questions remain open. However, the quantitative headline in the abstract ("up to 3 orders of magnitude") currently rests on extrapolated and synthetic evaluations rather than direct measurements, which weakens the empirical support for the claim as stated. The main contribution is the paradigm and the research agenda, which are valuable even if some of the reported numbers need to be recalibrated or replaced by more rigorous measurements.

major comments (5)
  1. [Section 4, Metrics observability] The "up to 3 orders of magnitude" claim in the abstract is supported in large part by the 48x/81x transmission/analytics reductions, which are obtained by "linearly extrapolat[ing] resource usage by 100x" from 1M to 100M time-series. If both the baseline and the ASAP pipeline scale linearly with the number of time-series, their ratio is invariant under this extrapolation, so the extrapolation cannot by itself produce the reported factors. The paper needs either direct measurements at the larger scale or an explicit scaling model with justification and error bars; otherwise the hyper-scale benefit is unsupported.
  2. [Section 4, Metrics observability] The reported 28–3000x query speedup and 49–7000x memory reduction compare sketch-based approximate answers to exact answers under a 95% relative accuracy target, but the paper does not report the actual per-query fidelity achieved. Without the empirical error distribution (e.g., the fraction of quantile, cardinality, and top-k queries that met the 95% target and the worst-case error), it is impossible to tell whether the speedups are achieved within the stated accuracy budget. Please report per-query relative errors and the accuracy-performance tradeoff.
  3. [Section 4, Distributed tracing] The 140x compression and 50x transmission reduction are measured after sampling 50K of 531K Uber traces and "remap[ping] them into a 1-second epoch" (footnote 2). Replacing the original multi-day randomized timestamps with a 1-second epoch removes the temporal spread that makes real trace data costly to compress, so the compression ratio may be inflated. The evaluation should either use the original timestamps or quantify the effect of the remap on compressibility.
  4. [Section 5, Cross-lifecycle primitive placement & configuration] The paper correctly identifies as open "ensuring that upstream reductions preserve the information required downstream." This is the property on which the cross-lifecycle benefits of ASAP ultimately depend: without a composition or preservation guarantee, the demonstrated gains for aggregate-query prototypes do not automatically extend to the broader workloads the paradigm targets. The paper should either provide a concrete example or formal condition under which compositions preserve downstream fidelity, or explicitly scope the headline claim to the aggregate-query setting.
  5. [Section 4, overall evaluation design] The three prototypes combine multiple primitives across stages, but there is no ablation separating the contribution of cross-lifecycle composition from the contribution of individual per-stage primitives. To support the claim that the unified view (rather than simply using a sketch at one stage) is what unlocks the gains, the evaluation should compare, for example, sketch-at-collector versus sketch-at-query-time, and single-primitive versus composed pipelines.
minor comments (5)
  1. [Abstract and Section 1] The acronym ASAP is introduced in both the abstract and the body with slightly different formatting; consider unifying the definition and use one canonical spelling throughout.
  2. [Figure 1] The right half of Figure 1 is dense and hard to parse, especially in monochrome; consider using labeled boxes or a table to show how the framework synthesizes bespoke pipelines from the primitives.
  3. [Section 4, Metrics observability] The experimental setup is underspecified: the paper does not state the machine configuration, sketch parameters (e.g., widths, depths), or the per-query configuration that produces the endpoints of the 49–7000x memory reduction range. Adding these details would improve reproducibility.
  4. [Section 4, Network monitoring] The network monitoring result reports a 600x storage reduction with "over 50% fidelity improvement" compared with the state of the art, but the fidelity metric and the baseline configuration are never defined, making the result hard to interpret.
  5. [Section 2, Table 1] Table 1 lists "Rollups" for business analytics and "Graph summarization" for observability tracing, but the later taxonomy in Section 3.1 classifies primitives by preserved semantics; a brief mapping between the table's entries and the taxonomy would clarify which primitive family each row belongs to.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; quantitative claims are measured prototype results against external baselines.

full rationale

The paper is a position paper whose central claim is a paradigm proposal, not a derived theorem. The reported CSP improvements are presented as measured outcomes of three implemented pipelines compared with external systems (Prometheus, ClickHouse) and public datasets (Uber, CAIDA). The metrics observability numbers come from a concrete evaluation setup with 1M time-series and query workloads; the tracing and network-monitoring numbers are similarly experimental. The self-cited artifacts (ASAPCollector, ASAPQuery, T-PACK) are components used in the evaluations, but the results are not assumed from those citations; they are benchmarked in the present paper. The linear extrapolation to 100M time-series is a scaling estimate, not a fitted parameter renamed as a prediction, and it is not definitionally forced. Finally, the paper explicitly acknowledges that cross-lifecycle fidelity-preserving composition is an open challenge (Section 5), so the framework does not claim to derive its benefits from a composition guarantee. No step in the paper's argument reduces, by construction or by self-citation, to its own inputs.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities; it reframes existing primitives under an umbrella abstraction. The load-bearing premises are the composability of primitives across stages, the representativeness of two transformed evaluation datasets, and a linear extrapolation to hyper-scale.

free parameters (1)
  • Hyper-scale extrapolation factor = 100
    The paper linearly extrapolates resource usage from 1M to 100M time-series by 100x to compute the 48x and 81x reductions in transmission and analytics cost; the central '3 orders of magnitude' claim for metrics depends on this assumption.
assumptions (4)
  • domain assumption Existing reduction primitives (sampling, sketching, wavelets, rollups, compression) preserve the application semantics needed for downstream queries.
    The paper's central taxonomy in Section 3.1 defines primitive families by the semantics they preserve. The assumption that these primitives are compositional across lifecycle stages is explicitly listed as an open challenge in Section 5.
  • ad hoc to paper Linear extrapolation of resource usage from 1M to 100M time-series by a factor of 100x yields a valid upper bound on hyper-scale benefits.
    In Section 4 (metrics observability), the 48x and 81x reductions in transmission and analytics cost are obtained by linear extrapolation, not measurement. If resource scaling is not linear, the 3-orders-of-magnitude claim for that domain is unsupported.
  • ad hoc to paper Sampling 50K of 531K Uber traces and remapping them into a 1-second epoch produces a workload representative of production distributed tracing.
    Footnote 2 in Section 4. This transformation discards original temporal structure and may not reflect real-world trace arrival patterns, which is load-bearing for the tracing results.
  • domain assumption The 95% relative accuracy target is an acceptable fidelity requirement for the applications.
    Section 4 evaluation; chosen by the authors, not derived from application requirements.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ASAP: Reimagining the Data Lifecycle using Application Semantic-Aware Processing." pith.science (2026). https://pith.science/paper/3IWT3IEO

@misc{pith2026260812735,
  author       = {Pith},
  title        = {Pith review of: ASAP: Reimagining the Data Lifecycle using Application Semantic-Aware Processing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3IWT3IEO}},
  note         = {Machine review of arXiv:2608.12735}
}
read the original abstract

Across many domains (e.g., observability, networking, security), data processing pipelines face what we refer to as the CSP problem: achieving low Cost at large Scale, while maintaining high Performance. In response, we see several efforts to tackle CSP in various stages of the Collect-Transmit-Store-Analyze data lifecycle; such as approximate query processing in databases or sketches in network routers. Our work is driven by the simple insight: "seeing the forest for the trees". These proposed solutions (e.g., AQP, sketching, compression, rollups) addressing CSP share a common property - they exploit semantic-preserving opportunities to support application needs. In this paper, we make a case for ASAP, a paradigm that makes Application Semantic-Aware Processing (ASAP) a first-class design principle in data processing pipelines. We argue that by taking a unified view across ASAP primitives developed in different domains, across the entire data lifecycle, we can unlock new opportunities to tackle the CSP problem. In particular, we can: (i) enable novel cross-lifecycle optimizations such as analytics run directly on sketches computed at the source; (ii) leverage primitives developed in other application domains; and (iii) enable widespread adoption of these powerful techniques. We discuss research challenges in socializing the benefits of the ASAP paradigm, and show preliminary evidence that adopting ASAP can yield up to 3 orders of magnitude improvements in the CSP tradeoff for many application domains.

Figures

Figures reproduced from arXiv: 2608.12735 by the authors.

Figure 1
Figure 1. ASAP makes Application Semantic-Aware Processing primitives first-class citizens to enable cross-domain and cross-lifecycle benefits. 2. Scale: Data volume is growing because deployments are larger and more complex, and because operators require finer-grained signal data to understand these systems [23]. Critically, this growth is no longer just human-scale. Machine￾generated traffic, from bots and AI agents [20, 42… view at source ↗
Figure 2
Figure 2. Three implemented ASAP-based pipelines to show early promise of the ASAP paradigm. ASAPTransmit Collect Transmit Store Analyze ASAP Traditional (Raw data) OTel Collector OTLP protocol Filesystem store Prometheus ASAPCollect ASAPStore ASAPQuery [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. ASAP pipeline for metrics observability. For example, in a metrics observability pipeline, an opera￾tor may need to support tail-latency and heavy-hitter queries while facing high network and storage costs. The deploy￾ment specification may further require compatibility with an existing Prometheus backend and impose a strict CPU budget at collectors. Based on these specifications, the ASAP framework may select quant… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 43 canonical work pages

  1. [1]

    https://hotcarbon.org/

    HotCarbon Workshop on Sustainable Computer Systems. https://hotcarbon.org/

    work page

  2. [2]

    https://catalog.caida.org/dataset/ passive_2018_pcap

    Anonymized Internet Traces 2018, 2018. https://catalog.caida.org/dataset/ passive_2018_pcap

    work page 2018

  3. [3]

    https://clickhouse.com/

    Clickhouse, 2026. https://clickhouse.com/

    work page 2026

  4. [4]

    https://docs.cloud.google.com/bigquery/ docs/reference/standard-sql/approximate_aggregate_functions

    Approximate aggregate functions, 2026. https://docs.cloud.google.com/bigquery/ docs/reference/standard-sql/approximate_aggregate_functions

    work page 2026

  5. [5]

    https: //github.com/ProjectASAP/ASAPCollector-public

    ASAPCollector: Reducing Metrics Observability Transmission Costs, 2026. https: //github.com/ProjectASAP/ASAPCollector-public

    work page 2026

  6. [6]

    https://github

    ASAPQuery: Drop-in Acceleration of SQL/PromQL queries, 2026. https://github. com/ProjectASAP/ASAPQuery

    work page 2026

  7. [7]

    https://grafana.com/ blog/the-red-method-how-to-instrument-your-services/

    The RED Method: How to Instrument Your Services, 2026. https://grafana.com/ blog/the-red-method-how-to-instrument-your-services/

    work page 2026

  8. [8]

    https://grafana.com/observability- survey/2025/

    Observability Survey Report 2025, 2026. https://grafana.com/observability- survey/2025/

    work page 2025

Show all 55 references
  1. [9]

    https://opentelemetry.io/

    OpenTelemetry, 2026. https://opentelemetry.io/

    2026

  2. [10]

    https://pcpartpicker.com/trends/price/memory

    Memory Price Trends, 2026. https://pcpartpicker.com/trends/price/memory

    2026

  3. [11]

    https://prometheus.io/

    Prometheus, 2026. https://prometheus.io/

    2026

  4. [12]

    https://thanos.io/

    Thanos, 2026. https://thanos.io/

    2026

  5. [13]

    https://www.honeycomb.io/ blog/how-much-should-i-spend-on-observability-pt1

    How Much Should I Spend on Observability?, 2026. https://www.honeycomb.io/ blog/how-much-should-i-spend-on-observability-pt1

    2026

  6. [14]

    https://www.iea.org/reports/energy-and-ai/ energy-demand-from-ai

    Energy demand from AI, 2026. https://www.iea.org/reports/energy-and-ai/ energy-demand-from-ai

    2026

  7. [15]

    Why Observability Budgets Keep Growing Even When IT is Asked to Cut Costs,

  8. [16]

    Agarwal, B

    S. Agarwal, B. Mozafari, A. Panda, H. Milner, S. Madden, and I. Stoica. Blinkdb: queries with bounded errors and bounded response times on very large data. In Proceedings of the 8th ACM European Conference on Computer Systems, EuroSys ’13, page 29–42, New York, NY, USA, 2013. ...

    2013

  9. [17]

    Anderson, R

    T. Anderson, R. Mahajan, S. Peter, and L. Zettlemoyer. Self-defining systems,

  10. [18]

    Cheng, S

    A. Cheng, S. Liu, M. Pan, Z. Li, B. Wang, A. Krentsel, T. Xia, M. Cemri, J. Park, S. Yang, J. Chen, L. Agrawal, A. Desai, J. Xing, K. Sen, M. Zaharia, and I. Stoica. Barbarians at the gate: How ai is upending systems research, 2025. https://arxiv. org/abs/2510.06189

    2025

  11. [19]

    Y.-R. Chin, M. Srivastava, Y. Zhou, G. Fanti, and V. Sekar. T-pack: Structure-aware generative compression for efficient distributed tracing. InTo appear at NSDI 2027, 2027

    2027

  12. [20]

    AI and bots have officially taken over the internet, 2026

    CNBC. AI and bots have officially taken over the internet, 2026. https://www. cnbc.com/2026/03/26/ai-bots-humans-internet.html

    2026

  13. [21]

    Cormode, M

    G. Cormode, M. Garofalakis, P. J. Haas, and C. Jermaine. Synopses for mas- sive data: Samples, histograms, wavelets, sketches.Foundations and Trends in Databases, 4(1-3):1–294, 2011

    2011

  14. [22]

    Cornacchia, T

    A. Cornacchia, T. A. Benson, M. Bilal, and M. Canini. Observability is eating your cores: Fine-grained analysis of microservice metrics with ipu-hosted sketches. 2026

    2026

  15. [23]

    Dean and L

    J. Dean and L. A. Barroso. The tail at scale.Communications of the ACM, 56(2):74– 80, 2013

    2013

  16. [24]

    F. Dong, Y. Yin, S. Liu, and V. Sekar. Tackling long-term network trace retention challenges using deep generative compression. 2025

    2025

  17. [25]

    Eriksen, K

    M. Eriksen, K. Veeraraghavan, Y. Abdulghani, A. Birchall, P.-Y. Chou, R. Cornew, A. Kabiljo, M. Lieuw, J. Meza, S. Michelson, et al. Global capacity manage- ment with flux. In17th USENIX Symposium on Operating Systems Design and Implementation (OSDI 23), pages 589–606, 2023

    2023

  18. [26]

    P. Fei, Z. Li, Z. Wang, X. Yu, D. Li, and K. Jee.{SEAL}: Storage-efficient causality analysis on enterprise logs with query-friendly compression. In30th USENIX security symposium (USENIX Security 21), pages 2987–3004, 2021

    2021

  19. [27]

    D. Feldman. Introduction to core-sets: an updated survey, 2020. https://arxiv. org/abs/2011.09384

    2020 arXiv

  20. [28]

    Y. Gan, M. Liang, S. Dev, D. Lo, and C. Delimitrou. Sage: practical and scalable ml- driven performance debugging in microservices. InProceedings of the 26th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, pages 135–151, 2021

    2021

  21. [29]

    Ghotikar

    T. Ghotikar. AI Agent Workloads: PostgreSQL vs CockroachDB, 2026. https: //www.cockroachlabs.com/blog/scaling-ai-agents-database-concurrency

    2026

  22. [30]

    Grayeli, A

    A. Grayeli, A. Sehgal, O. Costilla-Reyes, M. Cranmer, and S. Chaudhuri. Symbolic regression with a learned concept library. InAdvances in Neural Information Processing Systems 37 (NeurIPS), 2024

    2024

  23. [31]

    Hamadanian, P

    P. Hamadanian, P. Karimi, A. Nasr-Esfahany, K. Noorbakhsh, J. Chandler, A. ParandehGheibi, M. Alizadeh, and H. Balakrishnan. Glia: A human-inspired ai for automated systems design and optimization. InProceedings of the ACM Con- ference on AI and Agentic Systems, CAIS ’26, page...

    2026

  24. [32]

    Count-Min Sketch: The Art and Science of Estimating Stuff,

    Itamar Haber. Count-Min Sketch: The Art and Science of Estimating Stuff,

  25. [33]

    Kamarthi, H

    H. Kamarthi, H. Shah, H. Milner, S. Sinha, Y. Li, B. A. Prakash, and V. Sekar. Aha: Scalable alternative history analysis for operational timeseries applications. Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1, 2026. https://api.semant...

    2026

  26. [34]

    S. Kim, C. Jung, R. Jang, D. Mohaisen, and D. H. Nyang. A robust counting sketch for data plane intrusion detection. In30th Annual Network and Distributed System Security Symposium, NDSS 2023. The Internet Society, 2023

    2023

  27. [35]

    I.-T. A. Lee, Z. Zhang, A. Parwal, and M. Chabbi. The tale of errors in mi- croservices.Proc. ACM Meas. Anal. Comput. Syst., 8(3), Dec. 2024. https: //doi.org/10.1145/3700436

    2024 doi

  28. [36]

    Z. Li, Q. Cheng, K. Hsieh, Y. Dang, P. Huang, P. Singh, X. Yang, Q. Lin, Y. Wu, S. Levy, et al. Gandalf: An intelligent,{End-To-End} analytics service for safe deployment in{Large-Scale} cloud infrastructure. In17th USENIX Symposium on Networked Systems Design and Implementati...

    2020

  29. [37]

    S. Liu, S. Ponnapalli, S. Shankar, S. Zeighami, A. Zhu, S. Agarwal, R. Chen, S. Suwito, S. Yuan, I. Stoica, M. Zaharia, A. Cheung, N. Crooks, J. E. Gonzalez, and A. G. Parameswaran. Supporting our ai overlords: Redesigning data systems to be agent-first, 2025. https://arxiv.or...

    2025

  30. [38]

    Z. Liu, H. Namkung, G. Nikolaidis, R. Agarwal, D. Walker, and J. Rexford. Univ- mon: A universal monitoring framework for data plane. InProceedings of the ACM SIGCOMM Conference, 2016

    2016

  31. [39]

    Masson, J

    C. Masson, J. E. Rim, and H. K. Lee. Ddsketch: a fast and fully-mergeable quantile sketch with relative-error guarantees.Proc. VLDB Endow., 12(12):2195–2205, Aug

  32. [40]

    Moritz, D

    D. Moritz, D. Fisher, B. Ding, and C. Wang. Trust, but verify: Optimistic visu- alizations of approximate queries for exploring big data. InProceedings of the 2017 CHI Conference on Human Factors in Computing Systems, CHI ’17, page 2904–2915, New York, NY, USA, 2017. Associati...

    2017

  33. [41]

    Y. Park, B. Mozafari, J. Sorenson, and J. Wang. Verdictdb: Universalizing approx- imate query processing. InProceedings of the 2018 International Conference on Management of Data, SIGMOD ’18, page 1461–1476, New York, NY, USA, 2018. Association for Computing Machinery. https:/...

    2018

  34. [42]

    Pasupathy

    K. Pasupathy. How much does it cost to run ai agents at scale?, 2026. https: //atlan.com/know/ai-agent/cost-to-run-ai-agents-at-scale/

    2026

  35. [43]

    Rzadca, P

    K. Rzadca, P. Findeisen, J. Swiderski, P. Zych, P. Broniek, J. Kusmierek, P. Nowak, B. Strack, P. Witusowski, S. Hand, et al. Autopilot: workload autoscaling at google. Inproceedings of the fifteenth european conference on computer systems, pages 1–16, 2020

    2020

  36. [44]

    Y. Tang, D. Li, Z. Li, M. Zhang, K. Jee, X. Xiao, Z. Wu, J. Rhee, F. Xu, and Q. Li. Nodemerge: Template based efficient data reduction for big-data causality analysis. InProceedings of the 2018 ACM SIGSAC conference on computer and communications security, pages 1324–1337, 2018

    2018

  37. [45]

    Wehrstein, T

    J. Wehrstein, T. Eckmann, M. Jasny, and C. Binnig. Bespoke olap: Synthesizing workload-specific one-size-fits-one database engines, 2026. https://arxiv.org/abs/ 2603.02001

    2026 arXiv

  38. [46]

    Y. Yin, Z. Lin, M. Jin, G. Fanti, and V. Sekar. Practical gan-based synthetic ip header trace generation using netshare. InProceedings of the ACM SIGCOMM 2022 Conference, SIGCOMM ’22, pages 458–472, New York, NY, USA, 2022. Association for Computing Machinery. https://doi.org/...

    2022

  39. [47]

    Zhang, Z

    L. Zhang, Z. Xie, V. Anand, Y. Vigfusson, and J. Mace. The benefit of hindsight: Tracing Edge-Cases in distributed systems. In20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23), pages 321–339, Boston, MA, Apr. 2023. USENIX Association. https://www.u...

    2023

  40. [48]

    Zheng, C

    H. Zheng, C. Huang, X. Han, J. Zheng, X. Wang, C. Tian, W. Dou, and G. Chen. umon: Empowering microsecond-level network monitoring with wavelets. In Proceedings of the ACM SIGCOMM 2024 Conference, ACM SIGCOMM ’24, page 274–290, New York, NY, USA, 2024. Association for Computin...

    2024

  41. [49]

    Y. Zhou, F. Zhao, E. Wang, A. K. Coskun, D. Agrawal, A. El Abbadi, and Z. Liu. {PrvTel}: Lightweight models for private and accurate telemetry data retention. In23rd USENIX Symposium on Networked Systems Design and Implementation (NSDI 26), pages 1827–1843, 2026

    2026

  42. [50]

    Y. Zhu, T. Jin, S. Baziotis, C. Zhang, C. Mendis, and D. Kang. Pilotdb: Database- agnostic online approximate query processing with a priori error guarantees. Proc. ACM Manag. Data, 3(3), June 2025. https://doi.org/10.1145/3725335

    2025 doi

  43. [51]

    Z. Zhu, J. Chamberlain, K. Wu, D. Starobinski, and Z. Liu. Approximation-first timeseries query at scale.Proceedings of the VLDB Endowment, 18(8):2348–2361, 2025. 7

    2025

  44. [2019]

    https://doi.org/10.14778/3352063.3352135

  45. [2022]

    https://redis.io/blog/count-min-sketch-the-art-and-science-of-estimating- stuff/

  46. [2025]

    https://sds.cs.washington.edu/papers/sds.pdf

  47. [2026]

    https://www.logicmonitor.com/blog/observability-budget-trends

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.