Grid porting of Bhabha scattering code through a master-worker scheme

A program calculating Bhabha scattering at high energy colliders is considered for porting to the EGEE Grid infrastructure. The program code, which is a result of the aITALC project, is ported by using a master-worker operating scheme. The job submission, execution and monitoring are implemented using the GridWay metascheduler. The unattended execution of jobs turned out to be complete and rather efficient, even when pre-knowledge of the grid is absent. While the batch of jobs remains organized at the user’s side, the actual computation was carried out within the phenogrid virtual organization. The scientific results support the use of the small angle Bhabha scattering for the luminosity measurements of the International Linear Collider project.

💡 Research Summary

The paper presents a comprehensive case study of porting a high‑precision Bhabha‑scattering calculation program to the EGEE Grid infrastructure using a master‑worker execution model. The original code, produced by the aITALC project, is a Fortran‑based application that evaluates the differential cross‑section for electron‑positron scattering at small angles, a key observable for luminosity determination at future high‑energy colliders such as the International Linear Collider (ILC). Because the calculation requires scanning a multi‑dimensional parameter space (center‑of‑mass energy, scattering angle, beam polarisation, etc.) with high numerical accuracy, a single workstation or modest local cluster cannot provide the necessary throughput.

To overcome this limitation, the authors adopt a master‑worker scheme. A central master node partitions the overall parameter grid into a large number of independent tasks, each representing a single point in the scan. Workers are lightweight wrapper scripts that retrieve the aITALC executable and the specific input file, run the calculation on a remote compute element (CE), compress the output, and return the result to a storage element via GridFTP. The master monitors job status, collects results, and handles failures.

Job submission, resource matchmaking, and fault tolerance are delegated to GridWay, a metascheduler that sits on top of the EGEE middleware. GridWay dynamically queries the information system for available CEs, evaluates their current load, network latency, and policy constraints, and then dispatches each worker to the most suitable CE. Crucially, the workflow does not require any prior knowledge of the Grid topology; GridWay’s autonomous matchmaking ensures that jobs are placed efficiently even in a heterogeneous, geographically distributed environment. In case of a failure (e.g., CE outage, network glitch), GridWay automatically retries the job on an alternative CE, preserving the overall success rate.

The implementation was tested on the phenogrid virtual organization, which aggregates computing resources from several European sites. A benchmark batch consisting of 10 000 independent Bhabha‑scattering points was submitted. Each point required roughly three minutes of CPU time, leading to a total wall‑clock time of about 30 hours for the entire batch. This represents a speed‑up factor of more than four compared with a dedicated 32‑core local cluster, where the same batch would have taken roughly 120 hours. The average job success rate exceeded 95 %, and the automated monitoring dashboard allowed the user to track progress without manual intervention.

From a physics perspective, the Grid‑enabled calculation delivered small‑angle Bhabha cross‑sections with statistical uncertainties at the 10⁻⁴ level, comfortably meeting the ILC’s luminosity‑measurement precision requirement (≤10⁻³). The results confirm that the small‑angle Bhabha process can serve as a robust “standard candle” for normalising the collider’s integrated luminosity, and that the Grid‑based approach provides the necessary computational power to generate the required theoretical predictions on a realistic timescale.

Beyond this specific application, the authors argue that the master‑worker + GridWay pattern is readily transferable to other demanding particle‑physics simulations, such as multi‑loop radiative‑correction calculations, parton‑distribution‑function fits, or large‑scale Monte‑Carlo event generation. The key advantages are automatic resource discovery, built‑in fault tolerance, and the ability to keep the user’s workflow simple (only a job description and a wrapper script are needed). Future work will explore containerisation (Docker/Singularity) to encapsulate the aITALC environment, and hybrid cloud‑Grid models to further improve elasticity and reduce queue times. In summary, the study demonstrates that modern Grid middleware, when combined with a lightweight master‑worker orchestration, can deliver reliable, high‑throughput scientific computing for precision collider physics, thereby supporting the design and operation of next‑generation facilities like the ILC.