Overview
About
The portal
DLT Portal is a friendly interface to a research-grade Divisible Load Theory engine. Describe a set of worker machines and a divisible workload, and it computes and visualises schedules that minimise completion time (and, when you care, energy). It is aimed at researchers, students and practitioners exploring how distribution strategy, heterogeneity, memory limits and energy trade-offs shape parallel performance.
The engine
Under the hood is a self-contained C++20 library with a portfolio of solvers covering instant closed forms, fast heuristics, provably-optimal branch-and-bound, and mixed-integer programs (via HiGHS). It models single-port communication, per-processor startup, bandwidth and speed, memory limits, multi-installment delivery, and a four-state energy criterion. The portal talks to it through a thin C-ABI bridge, so the engine stays fully usable as a standalone library. The full solver list and usage reference are in the Library manual. The source code is openly available on GitHub.
Use cases
Practical implementations
DLT has been applied across a wide range of parallel and distributed computing domains. Representative application areas are listed below, each supported by peer-reviewed work in the bibliography.
Parallel image registration, video frame encoding, and large-image workload distribution across bus-connected workstation clusters. DLT provides closed-form load fractions that match the throughput of each worker, eliminating manual tuning.
Protein sequence alignment on mesh multiprocessors, genomic data processing, and radar signature search in flat-file databases. DLT schedules data delivery and computation jointly so that neither subsystem is idle while the other is busy.
Measurement scheduling and sensing workload allocation for data fusion in hierarchical sensor networks. Active-sleep cycles and multi-installment schemes reduce energy while meeting data-reporting deadlines.
Scheduling data-intensive applications across heterogeneous grid nodes and cloud virtual machines with bandwidth and cost constraints. Multi-source and multi-QoS formulations handle bursty demand and monetary budgets.
Video-on-demand serving, stream scheduling in cluster-based routers, and window-assisted video partitioning for caching. Divisible load framing captures the dominant data-movement cost and yields near-optimal buffer assignments.
Packet scheduling in network processors, fair link striping with FIFO delivery on heterogeneous channels, and pipeline latency modelling. DLT naturally expresses the interplay between communication bandwidth and processing throughput.
People
Team
