Scheduling;
Render farm;
Batch schedulers;
High performance computing;
Cluster management;
D O I:
10.1007/978-3-031-22698-4_11
中图分类号:
TP301 [理论、方法];
学科分类号:
081202 ;
摘要:
The visual effects (VFX) industry has for decades been actively developing and running job and resource management systems (JARMS) under the guise of "render farms". Rarely have we sought to share our work, stories, and problems beyond our niche bubble with the wider HPC community. With this paper we hope to break that cycle and introduce you to how we at Industrial Light & Magic (ILM) support our ever-changing diverse workloads at scale to produce high quality imagery for major motion pictures and television such as Jurassic World and The Mandalorian. We then share recent development efforts to prepare us for our biggest challenge yet: ABBA Voyage. These developments are the result of practical improvements to a production scheduler yielding higher throughput, reduced waste and makespan, and increased insight into internal metrics. We also briefly discuss how we developed tooling to aid in trace-based simulations to gain confidence in production upgrades. This is the printed edition featuring greyscale images and dashed line types in line graphs. The original digital edition retains the coloured images and lines, which some readers may find offer a greater definition and distinction between data points, for example.
机构:
Kadir Has Univ, Econ, Istanbul, Turkey
Kadir Has Univ, Istanbul, Turkey
Int Dev Econ Associates IDEAs, New Delhi, IndiaKadir Has Univ, Econ, Istanbul, Turkey