Nitro High Throughput is a powerful and simple task launching solution designed to accelerate the execution of short computing jobs in high-throughput computing (HTC) and service-oriented architecture (SOA) workloads. Operating as an independent product, Nitro can also integrate seamlessly with any HPC scheduler, including Moab, TORQUE, SLURM, PBS Pro, Platform LSF, and UGE. Nitro eliminates launch speed bottlenecks by scheduling once for a large set of jobs, rather than individually, significantly improving overall system efficiency.

Features

• 100x Faster Throughput: Achieves up to 500 tasks per second per core in optimized environments.
• Supports Massive Scale: Handles thousands to billions of tasks in HTC or SOA workloads.
• Scheduler Integration: Works with Moab, TORQUE, SLURM, PBS Pro, Platform LSF, and UGE.
• Staged Coordinator-Worker Architecture: Throughput increases with the number of cores allocated.
• Simple Syntax: One line equals one task; executable commands are easily defined.
• Task Conversion Support: Includes task-filter tools to reformat existing job syntax.
• Viewpoint Portal Integration: Streamlined submission and management when used with Moab or SLURM.
• Session-Based Execution: Nitro jobs build, activate, and tear down sessions automatically.
• Scalable Sessions: Each session can span hundreds of nodes, launching hundreds of thousands of tasks per second.
• High-Speed Message Bus: Efficient task delegation and execution across nodes.

Benefits

• Drastically Improved Launch Speeds: Reduces bottlenecks for short jobs, boosting overall system performance.
• Enhanced Efficiency: Even partial Nitro adoption yields significant gains in system throughput.
• Flexible Deployment: Operates independently or integrates with existing HPC environments.
• Minimal Management Overhead: Sessions are seamless and require no manual oversight.
• Scalable Performance: Proven ability to run over a billion tasks on a single server over a weekend.
• Optimized for Short Jobs: Ideal for serial or single-node parallel jobs running milliseconds to minutes.
• Cost-Effective: Maximizes resource utilization and reduces idle time.