CROWDio: A Practical Mobile Crowd Computing Framework with Developer-Oriented Design, Adaptive Scheduling, and Fault Resilience

Disumi Pathirana; Kutila Gunasekara; Lakshani Manamperi; Nipun Premarathna; Thiwanka Pathirana

arxiv: 2604.19363 · v1 · submitted 2026-04-21 · 💻 cs.DC

CROWDio: A Practical Mobile Crowd Computing Framework with Developer-Oriented Design, Adaptive Scheduling, and Fault Resilience

Lakshani Manamperi , Disumi Pathirana , Thiwanka Pathirana , Nipun Premarathna , Kutila Gunasekara This is my paper

Pith reviewed 2026-05-10 01:43 UTC · model grok-4.3

classification 💻 cs.DC

keywords mobile crowd computingadaptive schedulingfault tolerancecheckpointingdeclarative SDKheterogeneous devicesdistributed computingAndroid

0 comments

The pith

CROWDio provides a declarative SDK, tiered checkpointing and telemetry-driven scheduling for practical mobile crowd computing on heterogeneous devices.

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

The paper presents CROWDio as a centralized framework that makes mobile crowd computing viable by letting developers distribute tasks through one function annotation while the system handles device differences, failures and connection issues automatically. It integrates a declarative interface for simplicity, a tiered checkpointing approach for resuming interrupted tasks on phones, and a scheduling system that adapts using live device data. A sympathetic reader would care because idle smartphone capacity could then support large-scale distributed work without requiring expert programmers or custom hardware. Evaluation on six varied Android devices shows the adaptive approach cuts total run time by up to 56.9 percent versus basic round-robin dispatch, keeps checkpoint costs to 2-3 seconds per task, and maintains fair workload sharing with a Jain index of 0.889.

Core claim

CROWDio is a centralized MCdC platform with three integrated subsystems: a declarative SDK that abstracts distributed execution to a single function annotation, a tiered checkpointing mechanism for fault-tolerant resumption under mobile memory and runtime limits, and a pluggable multi-criteria scheduling framework that uses continuous live device telemetry to support interchangeable decision strategies. On CPU-bound, AI/NLP and data-parallel workloads across six heterogeneous Android devices, capability-aware adaptive scheduling reduces total execution time by up to 56.9 percent relative to naive round-robin while the checkpointing subsystem adds only 2-3 seconds overhead per task regardless

What carries the argument

the pluggable multi-criteria scheduling framework driven by continuous live device telemetry, which selects devices according to measured capabilities to adapt dispatch decisions

If this is right

Developers distribute tasks using only a single function annotation without writing explicit parallelism or device-management code.
Tasks resume after faults via tiered checkpointing with bounded 2-3 second overhead independent of how often checkpoints occur.
Capability-aware scheduling using live telemetry outperforms round-robin dispatch by up to 56.9 percent in total execution time for CPU, AI inference and data-parallel workloads.
Workload distribution across heterogeneous devices achieves a system-wide Jain's Fairness Index of 0.889, indicating equitable and stable allocation.
The scheduling core accepts interchangeable decision modules so new strategies can be added without altering the dispatch logic.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The low and frequency-independent checkpoint overhead suggests the system could tolerate more frequent saves in highly volatile mobile environments without large performance penalties.
The declarative SDK abstraction could lower the entry barrier for integrating crowd-computing resources into ordinary mobile applications.
High fairness under adaptive scheduling implies the approach may sustain voluntary participation when users contribute spare device capacity.

Load-bearing premise

The six-device heterogeneous Android testbed and workloads accurately represent real-world device variability, connectivity volatility, and user behavior that the framework must handle at scale.

What would settle it

Deploying the same workloads on at least twenty additional devices under uncontrolled real-world networks with simulated interruptions and checking whether execution-time savings remain near 50 percent and checkpoint overhead stays under five seconds per task.

Figures

Figures reproduced from arXiv: 2604.19363 by Disumi Pathirana, Kutila Gunasekara, Lakshani Manamperi, Nipun Premarathna, Thiwanka Pathirana.

**Figure 2.** Figure 2: Worker disconnection and graceful reconnection lifecycle. Tasks resume on alter [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Execution time: best device, worst device, and CROWDio WRR across six workers. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

read the original abstract

Mobile Crowd Computing (MCdC) leverages the idle computational capacity of consumer smartphones to enable distributed task processing at scale; however, widespread real-world adoption remains constrained by the absence of developer-oriented frameworks capable of transparently managing device heterogeneity, fault tolerance, and connectivity volatility. This paper introduces CROWDio, a centralized MCdC platform comprising three tightly integrated subsystems: (i) a declarative SDK that abstracts distributed execution to a single function annotation, eliminating the need for explicit parallelism management; (ii) a tiered checkpointing mechanism that enables fault-tolerant task resumption under the memory and execution constraints inherent to mobile runtimes; and (iii) a pluggable multi-criteria scheduling framework driven by continuous live device telemetry, supporting interchangeable decision strategies without modification to the dispatch core. Empirical evaluation across six heterogeneous Android devices spanning CPU-bound, AI/NLP inference, and data-parallel workloads demonstrates that capability-aware adaptive scheduling reduces total execution time by up to 56.9% relative to naive round-robin dispatch, while the checkpointing subsystem incurs a bounded overhead of only 2-3 s per task regardless of checkpoint frequency. A system-wide Jain's Fairness Index of 0.889 confirms equitable and stable workload distribution across heterogeneous worker devices.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

CROWDio bundles a declarative SDK, tiered checkpointing, and telemetry-driven scheduling into one mobile crowd platform and shows concrete gains on small tests, but the six-device evaluation leaves scalability and real-world resilience unproven.

read the letter

The main takeaway is that this paper delivers a working framework that hides most of the distribution details behind a single function annotation, adds checkpointing that stays cheap on phones, and swaps in different schedulers based on live device stats. Those pieces together look like the actual advance over earlier mobile crowd work. The reported 56.9 percent drop in total run time versus round-robin and the 0.889 fairness index are the clearest evidence they provide that the adaptive part helps on heterogeneous hardware. The bounded 2-3 second checkpoint cost is also useful to see, since it suggests the fault-tolerance layer does not kill performance. They tested across CPU, inference, and data-parallel jobs, which gives the numbers a bit more weight than a single workload would. The soft spot is exactly what the stress-test note flags: everything rests on six Android devices. That setup cannot speak to hundreds of phones with frequent drops, changing networks, or non-stationary users, and the abstract gives no larger simulation, trace replay, or analytical bound to bridge the gap. Without those, the claims about practical fault resilience and scale stay speculative. Experimental controls, device selection rules, and any statistical checks are also not visible, so the numbers are hard to judge for robustness. This is for systems people who build or evaluate mobile and edge frameworks rather than theorists. A reader who needs an implemented example with developer abstractions and measurable scheduling trade-offs will find usable ideas here. It has enough concrete design and results to merit peer review, though the referees will almost certainly ask for bigger experiments and clearer methodology. I would send it forward with that expectation.

Referee Report

2 major / 2 minor

Summary. The paper introduces CROWDio, a centralized mobile crowd computing (MCdC) framework consisting of a declarative SDK for single-annotation distributed execution, a tiered checkpointing system for fault-tolerant resumption on mobile devices, and a pluggable multi-criteria scheduler using live device telemetry. On a six-device heterogeneous Android testbed running CPU-bound, AI/NLP, and data-parallel workloads, it reports up to 56.9% reduction in total execution time versus round-robin dispatch, 2-3 s bounded checkpoint overhead independent of frequency, and a system-wide Jain's Fairness Index of 0.889.

Significance. If validated at larger scale, the integrated developer-oriented design, adaptive scheduling, and bounded-overhead checkpointing would represent a practical advance for MCdC by lowering barriers to heterogeneous mobile task distribution while providing measurable efficiency and fairness gains. The pluggable scheduler and declarative abstraction are clear strengths that could enable broader adoption if the empirical claims are shown to generalize beyond the reported testbed.

major comments (2)

[§5] §5 (Experimental Evaluation): The headline claims of 56.9% execution-time reduction, 2-3 s checkpoint overhead, and JFI 0.889 rest entirely on a six-device heterogeneous Android testbed. No larger-scale simulation, trace-driven evaluation, or analytical model is provided to support extrapolation to hundreds of devices with realistic churn, variable connectivity, and non-stationary participation, which directly undermines the abstract's assertions of practicality 'at scale' and fault resilience under volatility.
[§5.1 and §5.2] §5.1 and §5.2: The reported performance numbers lack any description of experimental controls, device selection criteria, workload definitions, statistical significance testing, or variance across runs. This absence makes it impossible to assess whether the observed gains are robust or sensitive to the specific six-device configuration, rendering the central empirical support for the adaptive scheduler and checkpointing subsystems incomplete.

minor comments (2)

[Abstract and §1] The abstract and introduction use 'at scale' without a precise definition or scaling target; a brief clarification of the intended deployment regime would improve precision.
[§5] Figure captions and table headers in the evaluation section could more explicitly state the exact workload parameters and device models used for each data point.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the experimental evaluation. We agree that the current testbed size and lack of methodological details limit the strength of our scalability and robustness claims. We will revise the manuscript to incorporate additional simulation results and expanded methodology descriptions.

read point-by-point responses

Referee: [§5] §5 (Experimental Evaluation): The headline claims of 56.9% execution-time reduction, 2-3 s checkpoint overhead, and JFI 0.889 rest entirely on a six-device heterogeneous Android testbed. No larger-scale simulation, trace-driven evaluation, or analytical model is provided to support extrapolation to hundreds of devices with realistic churn, variable connectivity, and non-stationary participation, which directly undermines the abstract's assertions of practicality 'at scale' and fault resilience under volatility.

Authors: We acknowledge that the six-device real-world testbed does not by itself substantiate claims of practicality at the scale of hundreds of devices with churn. The evaluation was designed to validate the framework on heterogeneous Android hardware under realistic constraints rather than relying solely on simulation. In the revised manuscript, we will add a new subsection with trace-driven simulations based on publicly available mobile device participation traces, modeling up to 200 devices with variable connectivity and churn. This will provide quantitative support for extrapolation while preserving the emphasis on the integrated design contributions. revision: yes
Referee: [§5.1 and §5.2] §5.1 and §5.2: The reported performance numbers lack any description of experimental controls, device selection criteria, workload definitions, statistical significance testing, or variance across runs. This absence makes it impossible to assess whether the observed gains are robust or sensitive to the specific six-device configuration, rendering the central empirical support for the adaptive scheduler and checkpointing subsystems incomplete.

Authors: We agree that the experimental methodology in §§5.1 and 5.2 was insufficiently detailed. In the revision we will expand these sections to specify: device selection criteria (exact Android models, CPU/GPU/RAM specifications, and rationale for heterogeneity), workload definitions (precise input sizes and task parameters for CPU-bound, AI/NLP, and data-parallel benchmarks), experimental controls (network conditions, background app restrictions, and warm-up procedures), statistical methods (averaging over 10 independent runs per configuration with reported standard deviations and significance testing via paired t-tests), and variance analysis across runs. These additions will allow readers to evaluate result robustness directly. revision: yes

Circularity Check

0 steps flagged

No circularity: results are direct empirical measurements with no derivation chain or fitted inputs

full rationale

The paper describes a practical MCdC framework (SDK, tiered checkpointing, pluggable scheduler) and reports performance numbers (56.9% execution-time reduction, 2-3 s checkpoint overhead, JFI 0.889) as outcomes of experiments on a six-device Android testbed. No equations, parameters fitted to subsets of data, predictions derived from the framework's own definitions, or self-citations invoked as uniqueness theorems appear in the provided text. The central claims rest on observed measurements rather than any reduction to inputs by construction, satisfying the default expectation that the work is self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical axioms, free parameters, or invented physical entities; the paper is a systems-engineering contribution whose claims rest on engineering assumptions about mobile runtimes and device telemetry.

pith-pipeline@v0.9.0 · 5550 in / 1075 out tokens · 32385 ms · 2026-05-10T01:43:20.961891+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

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

Memory-Efficient Partitioned DNN Inference on Resource-Constrained Android Crowds
cs.LG 2026-05 unverdicted novelty 3.0

CROWDio enables memory-efficient ONNX inference of DistilBERT on Android handsets by partitioning across devices with JIT loading, affinity scheduling, compressed transport and streaming, keeping per-device memory at ...

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages · cited by 1 Pith paper

[1]

Pramanik, S

P.K.D. Pramanik, S. Pal, and P. Choudhury. Mobile crowd computing: potential, architecture, requirements, challenges, and applications.Journal of Supercomputing, 80(2), 2024.https://doi.org/10.1007/s11227-023-05545-0

work page doi:10.1007/s11227-023-05545-0 2024
[2]

Pramanik

P.K.D. Pramanik. Sustainable computing with mobile crowd computing. Technical Report, NIT Durgapur, 2023

work page 2023
[3]

Marinelli

E.E. Marinelli. Hyrax: Cloud computing on mobile devices using MapReduce. Technical Report CMU-CS-09-164, Carnegie Mellon University, 2009

work page 2009
[4]

Dou et al

A. Dou et al. Misco: A MapReduce framework for mobile systems. InProceedings of PETRA 2010. ACM, 2010

work page 2010
[5]

Fernando, S.W

N. Fernando, S.W. Loke, and W. Rahayu. Honeybee: A programming framework for mobile crowd computing. InMobiQuitous 2012, Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, vol. 120. Springer, 2013

work page 2012
[6]

Kiridana et al

T.B. Kiridana et al. Mobile crowd computing with mobile agents and work stealing. In Proceedings of MERCon 2024. IEEE, 2024

work page 2024
[7]

Transportation Research Part B: Methodological 46, 1159–1176

N. Fernando, S.W. Loke, and W. Rahayu. Mobile cloud computing: A survey.Fu- ture Generation Computer Systems, 29(1):84–106, 2013. https://doi.org/10.1016/j. future.2012.05.023

work page doi:10.1016/j 2013
[8]

A. Ray, C. Chowdhury, S. Bhattacharya, and S. Roy. A survey of mobile crowdsens- ing and crowdsourcing strategies for smart mobile device users.CCF Transactions on Pervasive Computing and Interaction, 5:98–123, 2023. https://doi.org/10.1007/ s42486-022-00110-9

work page 2023
[9]

Anderson

D.P. Anderson. BOINC: A platform for volunteer computing.Journal of Grid Computing, 18(1):99–122, 2020.https://doi.org/10.1007/s10723-019-09497-9

work page doi:10.1007/s10723-019-09497-9 2020
[10]

Saleh and C

E. Saleh and C. Shastry. A new approach for global task scheduling in volunteer computing systems.International Journal of Information Technology, 2022. https: //doi.org/10.1007/s41870-022-01090-w

work page doi:10.1007/s41870-022-01090-w 2022
[11]

Pramanik et al

P.K.D. Pramanik et al. A comparative analysis of MCDM methods for resource selection in MCC.Symmetry, 13(9):1713, 2021

work page 2021
[12]

C.E. Shannon. A mathematical theory of communication.Bell System Technical Journal, 27(3):379–423, 1948

work page 1948
[13]

R. Stan, C. Copil, D. Moldovan, and H.-L. Truong. Evaluation of task scheduling algorithms in heterogeneous computing environments.Sensors, 21(18):6035, 2021. https: //doi.org/10.3390/s21186035 9

work page doi:10.3390/s21186035 2021
[14]

Moritz et al

P. Moritz et al. Ray: A distributed framework for emerging AI applications. InProceedings of USENIX OSDI, pages 561–577, 2018

work page 2018
[15]

Zaharia et al

M. Zaharia et al. Apache Spark: A unified engine for big data processing.Communications of the ACM, 59(11):56–65, 2016

work page 2016
[16]

Fette and A

I. Fette and A. Melnikov. The WebSocket protocol (RFC 6455). IETF, 2011. https: //doi.org/10.17487/RFC6455

work page doi:10.17487/rfc6455 2011
[17]

Abdellatif

M. Abdellatif. Help desk tickets. Mendeley Data V2, 2025. https://doi.org/10.17632/ btm76zndnt.2

work page 2025
[18]

Bhathena

J. Bhathena. Weather image recognition dataset. Kaggle, n.d

work page
[19]

Keshavarz Ghorabaee et al

M. Keshavarz Ghorabaee et al. Multi-criteria inventory classification using EDAS.Infor- matica, 26(3):435–451, 2017

work page 2017
[20]

Zavadskas and Z

E.K. Zavadskas and Z. Turskis. A new ARAS method in multicriteria decision-making. Technological and Economic Development of Economy, 16(2):159–172, 2010

work page 2010
[21]

Pamucar and G

D. Pamucar and G. Cirovic. The selection of transport resources using MABAC.Expert Systems with Applications, 42(6):3016–3028, 2015. 10

work page 2015

[1] [1]

Pramanik, S

P.K.D. Pramanik, S. Pal, and P. Choudhury. Mobile crowd computing: potential, architecture, requirements, challenges, and applications.Journal of Supercomputing, 80(2), 2024.https://doi.org/10.1007/s11227-023-05545-0

work page doi:10.1007/s11227-023-05545-0 2024

[2] [2]

Pramanik

P.K.D. Pramanik. Sustainable computing with mobile crowd computing. Technical Report, NIT Durgapur, 2023

work page 2023

[3] [3]

Marinelli

E.E. Marinelli. Hyrax: Cloud computing on mobile devices using MapReduce. Technical Report CMU-CS-09-164, Carnegie Mellon University, 2009

work page 2009

[4] [4]

Dou et al

A. Dou et al. Misco: A MapReduce framework for mobile systems. InProceedings of PETRA 2010. ACM, 2010

work page 2010

[5] [5]

Fernando, S.W

N. Fernando, S.W. Loke, and W. Rahayu. Honeybee: A programming framework for mobile crowd computing. InMobiQuitous 2012, Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, vol. 120. Springer, 2013

work page 2012

[6] [6]

Kiridana et al

T.B. Kiridana et al. Mobile crowd computing with mobile agents and work stealing. In Proceedings of MERCon 2024. IEEE, 2024

work page 2024

[7] [7]

Transportation Research Part B: Methodological 46, 1159–1176

N. Fernando, S.W. Loke, and W. Rahayu. Mobile cloud computing: A survey.Fu- ture Generation Computer Systems, 29(1):84–106, 2013. https://doi.org/10.1016/j. future.2012.05.023

work page doi:10.1016/j 2013

[8] [8]

A. Ray, C. Chowdhury, S. Bhattacharya, and S. Roy. A survey of mobile crowdsens- ing and crowdsourcing strategies for smart mobile device users.CCF Transactions on Pervasive Computing and Interaction, 5:98–123, 2023. https://doi.org/10.1007/ s42486-022-00110-9

work page 2023

[9] [9]

Anderson

D.P. Anderson. BOINC: A platform for volunteer computing.Journal of Grid Computing, 18(1):99–122, 2020.https://doi.org/10.1007/s10723-019-09497-9

work page doi:10.1007/s10723-019-09497-9 2020

[10] [10]

Saleh and C

E. Saleh and C. Shastry. A new approach for global task scheduling in volunteer computing systems.International Journal of Information Technology, 2022. https: //doi.org/10.1007/s41870-022-01090-w

work page doi:10.1007/s41870-022-01090-w 2022

[11] [11]

Pramanik et al

P.K.D. Pramanik et al. A comparative analysis of MCDM methods for resource selection in MCC.Symmetry, 13(9):1713, 2021

work page 2021

[12] [12]

C.E. Shannon. A mathematical theory of communication.Bell System Technical Journal, 27(3):379–423, 1948

work page 1948

[13] [13]

R. Stan, C. Copil, D. Moldovan, and H.-L. Truong. Evaluation of task scheduling algorithms in heterogeneous computing environments.Sensors, 21(18):6035, 2021. https: //doi.org/10.3390/s21186035 9

work page doi:10.3390/s21186035 2021

[14] [14]

Moritz et al

P. Moritz et al. Ray: A distributed framework for emerging AI applications. InProceedings of USENIX OSDI, pages 561–577, 2018

work page 2018

[15] [15]

Zaharia et al

M. Zaharia et al. Apache Spark: A unified engine for big data processing.Communications of the ACM, 59(11):56–65, 2016

work page 2016

[16] [16]

Fette and A

I. Fette and A. Melnikov. The WebSocket protocol (RFC 6455). IETF, 2011. https: //doi.org/10.17487/RFC6455

work page doi:10.17487/rfc6455 2011

[17] [17]

Abdellatif

M. Abdellatif. Help desk tickets. Mendeley Data V2, 2025. https://doi.org/10.17632/ btm76zndnt.2

work page 2025

[18] [18]

Bhathena

J. Bhathena. Weather image recognition dataset. Kaggle, n.d

work page

[19] [19]

Keshavarz Ghorabaee et al

M. Keshavarz Ghorabaee et al. Multi-criteria inventory classification using EDAS.Infor- matica, 26(3):435–451, 2017

work page 2017

[20] [20]

Zavadskas and Z

E.K. Zavadskas and Z. Turskis. A new ARAS method in multicriteria decision-making. Technological and Economic Development of Economy, 16(2):159–172, 2010

work page 2010

[21] [21]

Pamucar and G

D. Pamucar and G. Cirovic. The selection of transport resources using MABAC.Expert Systems with Applications, 42(6):3016–3028, 2015. 10

work page 2015