What's New in Univa Grid Engine?

Version 8.0.1

<< Back to Univa Grid Engine product page

Improved Support for Multi-Core Hardware

• New HWLOC Library for Core Binding on Linux
• Applications run better on multi-core hardware.  More throughput and better Utilization

Integration with NVIDIA GPUs

• New CUDA Load Sensor for NVIDIA GPUs
• Allows inclusion of NVIDIA GPUs into a Univa Grid Engine Cluster.  Ensures GPU jobs run on the best (least loaded) GPU cores = More throughput in the cluster.  GPU jobs run more reliably
• The following metrics will be reported by this load sensor:
  • Total installed GPU memory
  • Total memory allocated
  • Total free memory
  • ECC Errors
  • Total GPU temperature
  • Power usage (in milliwatts)
• These additional metrics can be used for load thresholds or scheduling onto GPUs

New Job Submission Verifier Extensions

• Allows for better ‘seamless’ integration with customer scripts.  Provides an easier migration path for LSF customers

Additional Bug Fixes and improvements…more than 90!

• Improves overall stability of Univa Grid Engine.  Less downtime.

Univa UniSight 2.0 packaged with Univa Grid Engine

• Allows organizations to determine overall cluster utilization and throughput
• Data can be used to optimize hardware and software purchases
• New Analytics engine powered by Pentaho BI Suite(tm)
• Data Cubes allow customers to create ‘ad-hoc’ reports
• New reports can be created by Univa or any customer representative familiar with Pentaho

Additional Technical Improvements:

• Lots of Documentation Updates
• Client side JSV support ‘sync’
• Scheduler speedup for large reservations
• General cleanup and ‘refactoring’ of codebase
• Forwarding of some environment variables with interactive ‘qrsh’
• Qrshinteractive job enhancements
• Maximum array job instances and array job tasks
• Parallel jobs are not dispatched to least loaded host
• Updated qmonicons
• New fstype binary for detecting nfs4