{"id":"51e2913a-79b3-41a3-905c-8a63ca5fc16b","arxiv_id":"2510.02404","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A taxonomy and literature review of FaaS function resource configuration, organizing prior work by workload model, deployment environment, KPIs, resource targets, and configuration strategy.","lead":"This paper surveys research on how serverless 'functions' (small cloud programs) are given the right amount of memory, CPU, and other resources, and proposes a taxonomy of the factors that influence that choice. It categorizes about 40 existing studies by workload type, deployment platform, performance goals, resource target, and optimization strategy, and outlines open research gaps.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Survey's 'comprehensive' and 'first' claims rest on an unreported literature-selection process, and the classification tables show internal inconsistencies that suggest the taxonomy's empirical grounding is unreliable.","rationale":"The reader's conditional verdict is sound. I could not find a reason to reject the paper outright; the taxonomy is plausible and the prose is generally coherent. However, the weakest point is exactly where the reader pointed: the survey's evidentiary base. I went further and checked the internal consistency of the classification tables. The discrepancies (Wisefuse, Akhtar et al., and the non-survey [82]) are concrete and visible; they indicate the coding was not performed under a documented protocol. A systematic review without a protocol can still be useful as a narrative map, but its 'comprehensive' and 'first' claims require a stricter standard. My proposed test would settle this by checking whether the table can be reproduced and whether the search reveals missing literature. If the test passes (high agreement and no major omissions), the paper can be accepted as the first survey; if not, it should be revised to either narrow its claims or add a rigorous methodology. Therefore I recommend no change to the reader's CONDITIONAL verdict.","tokens_in":30863,"tokens_out":7874,"duration_ms":58788,"concrete_test":"Run a blinded re-coding study: have two independent researchers (not the authors) re-apply the paper's taxonomy to all 45 primary sources in Tables 2–3 using only the source abstracts/full texts, and simultaneously run a formal search (e.g., DBLP/Scopus: TITLE-ABS-KEY(('serverless' OR 'FaaS') AND ('function configuration' OR 'memory configuration' OR 'right-sizing'))) for 2018–2025. If the re-coding disagrees with Table 2/3 on >15% of cells, or the search returns >10 relevant peer-reviewed papers not cited (or any prior survey focused on function configuration), the 'comprehensive review' and taxonomy-as-map claims are not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"To support the central claims ('first survey' and 'comprehensive review' of function configuration), the paper needs (a) a reproducible literature search and (b) a consistent application of the taxonomy to the selected papers. Neither is currently demonstrable. The manuscript gives no search strategy, databases, or inclusion/exclusion criteria (§1, §3), so completeness is asserted rather than verified. The internal evidence is worse: the classification tables appear to contradict the prose. Wisefuse [55] is discussed in §3.1.1 as a workflow-characterisation work yet receives no ticks in Table 2. Akhtar et al. [70] are described in §3.4.1 as proposing a Bayesian-optimisation-based configuration method, but the corresponding Table 2 row does not mark Stochastic & Approximate Optimisation. Table 1 also lists [82]—a MICRO research paper, not a survey—among 'related surveys,' suggesting the selection of 'surveys' was ad hoc. These are not merely cosmetic: if the mapping of individual papers to taxonomy dimensions is unreliable, the taxonomy itself is not yet a dependable map of the area, and the 'first comprehensive' designation is unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper surveys research on resource configuration for Function-as-a-Service (FaaS) workloads. It proposes a taxonomy with five top-level dimensions—workload characteristics, deployment environment, key performance indicators, resource optimisation targets, and configuration/management strategy—and applies this taxonomy to classify roughly 40 papers from the serverless systems literature. The paper also compares prior serverless surveys and identifies future research directions. Its central claims are that it is the first survey focused specifically on function configuration and that it provides a comprehensive review of this area.","tokens_in":31174,"tokens_out":7871,"duration_ms":64713,"significance":"If the taxonomy and classification are reliable, the paper would be a useful contribution: it organizes a fragmented body of work on FaaS function configuration, gives a shared vocabulary for workload models, deployment environments, KPIs, resource targets, and solution strategies, and identifies concrete gaps for future work. The granular taxonomy is more focused than prior serverless resource-management surveys, and the paper collects many recent (2019–2024) systems papers that are otherwise scattered. However, the paper's current internal inconsistencies—most visibly in the classification tables—undermine the claim that the taxonomy is a dependable map of the area, and the absence of a documented literature-selection protocol prevents verification of the 'comprehensive' and 'first survey' claims. These issues must be addressed before the survey can serve as a reliable reference.","major_comments":[{"comment":"The paper presents itself as a 'comprehensive review' and 'the first survey that focus on the function configuration aspect of the FaaS model', but no systematic literature-search methodology is reported. There is no description of databases searched, query terms, inclusion/exclusion criteria, screening process, or quality assessment. For a survey whose central contribution is a taxonomy and a gap analysis, the literature set is load-bearing: without a reproducible protocol, completeness and representativeness cannot be verified. I recommend adding a methodology section (databases, time frame, search strings, inclusion/exclusion criteria, and ideally a PRISMA-style flow diagram), or softening the 'comprehensive'/'first survey' claims to match the actual narrative selection.","section":"Section 1, 3, 4"},{"comment":"As printed, the 'Our work' row does not mark the Configuration column as covered. The row reads: ✗ ✗ ✓ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗, placing the single ✓ under 'Workload General Overview' rather than under 'Configuration'. This directly contradicts Section 1's statement that this is the first survey focused on the function configuration aspect and the paper's title. The row must be corrected so that Configuration is marked covered; if the ✓ under Workload General Overview is also intended, that should be made explicit.","section":"Table 1, 'Our work' row"},{"comment":"The classification table is inconsistent with the prose. Wisefuse [55] is described in Section 3.1.1 as relevant to workflow/function-chain composition and in Section 3.3.3 as evidence that direct function-to-function communication is difficult, yet its Table 2 row is all ✗, including the 'Workflow/Function Chain' column. Similarly, Section 3.4.1 states that Akhtar et al. [70] 'propose a Bayesian Optimisation-based prediction method', but the Table 2 row for [70] does not mark 'Stochastic & Approximate Optimisation'. Whichever categorization is intended, the table and the text disagree. These mismatches suggest that the mapping of individual papers to taxonomy dimensions is not yet dependable; the tables should be regenerated from a consistent mapping and checked against the narrative.","section":"Table 2, rows [55] and [70]"},{"comment":"Table 1 is titled 'A categorisation of related surveys in Serverless Computing', but several entries are not surveys: [82] is a MICRO research paper on FaaS architecture, [73] is a preliminary study, [46] is an evaluation of production environments, and [91] is an empirical study of development challenges. Additionally, the row for [91] gives the year as 2017, while the reference list gives 2021. This makes the related-survey comparison unreliable and weakens the basis for the 'first survey' claim. The table should be restricted to actual surveys (or relabeled as 'related studies'), and the metadata should be corrected.","section":"Table 1, composition"}],"minor_comments":[{"comment":"Grammar: 'the first survey that focus' should be 'the first survey that focuses'.","section":"Section 1"},{"comment":"Bayesian optimisation is attributed to [70] in Section 3.4.1 but to [4] in Section 3.5.2. Please clarify which paper(s) actually use BO, and align the text and the classification table.","section":"Section 3.5.2 / 3.4.1"},{"comment":"The two table parts have the same caption 'A classification of resource management techniques' and appear on separate pages. Use a single table with a shared header or mark the second as a continuation.","section":"Table 2 / Table 3"},{"comment":"Typographical issues: 'server-less' in the abstract, 'sudy' for 'study' in Section 3.5.3, and 'A WS' for 'AWS' in several references and figure text.","section":"Throughout"},{"comment":"The year for [91] should be 2021 (per the reference list), not 2017 as shown in the table.","section":"Table 1"}],"recommendation":"major_revision","confidential_remarks":"The taxonomy is a plausible and potentially useful contribution, but the manuscript is not yet reliable as a survey: the missing methodology and the numerous table/prose inconsistencies directly affect the central 'first comprehensive survey' claim. These are fixable, so I would not recommend rejection, but the revision should be re-reviewed after the tables and methodology are corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is the first survey I know of that focuses specifically on function configuration in FaaS, and the taxonomy it proposes is coherent enough to be useful for positioning future work. But the paper's own tables don't always match its prose, and there is no description of how the literature was selected. Those problems are fixable, but they are exactly the kind of thing a referee should push on.\n\nWhat's genuinely new: the five-part taxonomy — workload model, deployment environment, KPIs, resource targets, and strategy — gives a sensible way to organize the roughly 40 papers it covers. I looked through several of the classifications and most are reasonable. The paper also does a decent job of distinguishing coupled vs. decoupled resource configuration and calling out gaps like configuration-aware scheduling and dynamic input-sensitive allocation. That is real value for someone entering the area.\n\nWhere it gets shaky: the 'first survey' and 'comprehensive review' claims rest on a literature-selection process that is not described. No search terms, no databases, no inclusion/exclusion criteria. That alone would be a minor complaint for a short survey, but here the selection is load-bearing: the taxonomy's dimensions are derived from the chosen papers, and the gap analysis is only as good as the coverage. Without the methodology, 'comprehensive' is an assertion, not a result.\n\nWorse, the internal evidence undercuts confidence in the classifications. In §3.1.1, Wisefuse [55] is described as a workflow-characterisation contribution, but its row in Table 2 is all crosses. Akhtar et al.'s work [70] is treated as Bayesian-optimisation-based in §3.4.1, but the corresponding row does not tick Stochastic & Approximate Optimisation. And Table 1 includes [82], a MICRO research paper, in a list of 'related surveys'. These are not just typos in the formatting; they suggest the mapping from papers to taxonomy dimensions was done carefully at the level of individual claims, and the table may not have been checked against the prose. Also, 'Akhtar et al. [70]' should be 'Raza et al.' based on the reference list.\n\nIf I were editing: send it to review. The taxonomy itself is a useful artifact and the domain is underserved by surveys. But the authors should be asked to add a methodology section, fix the table/prose conflicts, and re-check the reference attributions. With those changes, I'd cite it as a starting point.","headline":"A useful first map of FaaS function configuration, but the missing methodology and a few table/prose mismatches should be fixed before the classifications are treated as reliable.","tokens_in":31581,"tokens_out":4335,"would_cite":true,"duration_ms":29271,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This survey argues that serverless function resource configuration is a first-class research area and offers a five-dimension taxonomy of existing techniques, positioning itself as the first survey on this aspect of FaaS.","keywords":["serverless computing","Function-as-a-Service","function configuration","resource allocation","taxonomy","workload model","resource optimisation","research gaps"],"falsifier":"Check whether a keyword search across standard computing literature databases for phrases like 'serverless function configuration' or 'FaaS resource allocation' returns a comparable earlier survey focused specifically on configuration, or a paper that fits none of the taxonomy's five dimensions; either finding would undercut the 'first survey' claim or the taxonomy's exhaustiveness.","tokens_in":30807,"feed_emoji":"🧩","tokens_out":4323,"duration_ms":36551,"temperature":0.7,"pith_summary":"The paper sets out to make a research field out of a practical headache: when a developer deploys a serverless function, they must pick memory, CPU, timeout, and concurrency settings that balance cost and performance with almost no visibility into how the platform allocates resources. It argues that the many scattered proposals for solving this problem—prediction, search, machine learning, analytical models—can be organized by five dimensions: workload shape, deployment platform, performance metric, resource being tuned, and optimization strategy. The paper claims to be the first survey to focus on this specific aspect of FaaS, and its classification table plus gap analysis are meant as a map for researchers and a reference for developers and providers.","feed_headline":"One taxonomy maps forty studies on serverless function tuning","feed_subtitle":"The five-dimension frame shows where research clusters and where the open problems lie in function configuration.","key_machinery":"The taxonomy: a five-dimension classification scheme (workload model, deployment environment, key performance indicators, resource optimisation target, configuration strategy) that maps the surveyed literature into cells and exposes which combinations have been studied. It is the load-bearing structure of the paper; the gap analysis and future research directions are derived from cells that are empty or sparse.","core_discovery":"The paper's contribution is a classification, not a new algorithm. It claims that function resource configuration in serverless computing is a distinct research topic, and that the literature can be organized under a taxonomy of workload model (single function vs. workflow/function chain), deployment environment (commercial vs. open-source), key performance indicators (resource utilisation, lifecycle time and cost, data locality), resource optimisation target (coupled memory, decoupled CPU, application-specific), and management strategy (search and path-finding, stochastic/approximate optimisation, machine learning, formal/analytical). Reading roughly forty studies through this frame, the pa","pith_inferences":["The taxonomy may undercount work published in industry venues or adjacent fields such as serverless autoscaling and resource management that address configuration without using that exact term; those works could fit awkwardly or not fit at all.","If the platform landscape shifts toward decoupled resource allocation and specialized hardware, the 'resource optimisation target' dimension may need new sub-dimensions (GPU, FPGA, network) that the current survey does not list.","The observed dominance of memory-configuration studies likely reflects what commercial platforms expose rather than what matters most; a testable extension is whether decoupled configuration research grows as open-source frameworks and heterogeneous hardware become more central.","The 'first survey' framing depends on how 'configuration' is scoped; a related survey on general serverless resource management already exists, so the distinctiveness claim may be sharper than the field's boundaries truly are."],"forward_implications":["Researchers in serverless computing gain a shared vocabulary and a reference point for situating new work on function configuration.","The empty and sparse cells of the taxonomy identify concrete open problems: configuration-aware scheduling, dynamic runtime allocation, decoupled resource tuning, and workflow-level configuration.","The survey's claim to be the first on this aspect gives later surveys a baseline to compare against and a corpus to build on.","For developers and providers, the taxonomy shows how the coupled memory-based model of commercial platforms shapes what solutions are even possible.","The classification tables let a reader quickly see which combinations of workload type, platform, KPI, resource target, and strategy have been tried, and which have not."],"fun_headline_variants":["Five axes sort serverless function tuning studies","Taxonomy unpacks serverless function config puzzle","Forty studies, five lenses: serverless config map","Fresh frame for function resource choices in FaaS","Mapping the maze: serverless function configuration"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The whole taxonomy depends on the authors having chosen the right papers to review; they do not document a systematic search strategy or inclusion criteria, so if the corpus is skewed, the categories and the gap analysis are skewed with it.","fun_headline_variants_meta":{"raw":{"variants":["Five axes sort serverless function tuning studies","Taxonomy unpacks serverless function config puzzle","Forty studies, five lenses: serverless config map","Fresh frame for function resource choices in FaaS","Mapping the maze: serverless function configuration"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000201,"raw_usage":{"total_tokens":1228,"prompt_tokens":770,"completion_tokens":458,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":514,"completion_tokens_details":{"reasoning_tokens":387}},"tokens_in":514,"tokens_out":458,"duration_ms":4367,"temperature":1.0,"reasoning_tokens":387,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T12:51:09.905112+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check whether a keyword search across standard computing literature databases for phrases like 'serverless function configuration' or 'FaaS resource allocation' returns a comparable earlier survey focused specifically on configuration, or a paper that fits none of the taxonomy's five dimensions; either finding would undercut the 'first survey' claim or the taxonomy's exhaustiveness.","supporting_citations":[],"review_version":1}