A GPU Scheduling Framework to Accelerate Hyper-Parameter Optimization in Deep Learning Clusters

This paper proposes Hermes, a container-based preemptive GPU scheduling framework for accelerating hyper-parameter optimization in deep learning (DL) clusters. Hermes accelerates hyper-parameter optimization by time-sharing between DL jobs and prioritizing jobs with more promising hyper-parameter combinations. Hermes’s scheduling policy is grounded on the observation that good hyper-parameter combinations converge quickly in the early phases of training. By giving higher priority to fast-converging containers, Hermes’s GPU preemption mechanism can accelerate training. This enables users to find optimal hyper-parameters faster without losing the progress of a container. We have implemented Hermes over Kubernetes and compared its performance against existing scheduling frameworks. Experiments show that Hermes reduces the time for hyper-parameter optimization up to 4.04 times against previously proposed scheduling policies such as FIFO, round-robin (RR), and SLAQ, with minimal time-sharing overhead.

Processor Scheduling in High-Performance Computing (HPC) Environment

Processor scheduling is one of the thrust areas in the field of computer science. The future technologies use a huge amount of processing for execution of their tasks like huge games, programming software, and in the field of quantum computing. In real-time, many complex problems are solved by GPU programming. The primary concern of scheduling is to reduce the time complexity and manpower. Several traditional techniques exit for processor scheduling. The performance of traditional techniques is reduced when it comes to the huge processing of tasks. Most scheduling problems are NP-hard in nature. Many of the complex problems are recently solved by GPU programming. GPU scheduling is another complex issue as it runs thousands of threads in parallel and needs to be scheduled efficiently. For such large-scale scheduling problems, the performance of state-of-the-art algorithms is very poor. It is observed that evolutionary and genetic-based algorithms exhibit better performance for large-scale combinatorial and internet of things (IoT) problems.

Multi-Processor Job Scheduling in High-Performance Computing (HPC) Systems

Processor scheduling is one of the thrust areas in the field of computer science. The future technologies use a huge amount of processors for execution of their tasks like huge games, programming software, and in the field of quantum computing. In hard real-time, many complex problems are solved by GPU programming. The primary concern of scheduling is to reduce the time complexity and manpower. There are several traditional techniques for processor scheduling. The performance of traditional techniques is reduced when it comes under huge processing of tasks. Most scheduling problems are NP-hard in nature. Many of the complex problems are recently solved by the GPU programming. GPU scheduling is another complex issue as it runs thousands of threads in parallel and needs to be scheduled efficiently. For such large-scale scheduling problem, the performance of state-of-the-art algorithms is very poor. It is observed that evolutionary and genetic-based algorithms exhibit better performance for large-scale combinatorial problems.