Univa® Unisight™ is uniquely crafted to make visible the metrics that characterize objects specific to Univa® Grid Engine® such as jobs, plus SUSE Linux hosts, clusters and clouds. End users and admins can choose to make use of pre-existing tables and charts, based on real-time or historical monitoring and reporting metrics, or develop their own. As of the latest Unisight release, however, all of these metrics have become actionable through an enhanced Rules Engine. Beyond alerting, metrics known to Unisight can be framed as variables, and passed to various handlers - e.g., display, email, scripting or even to another rule. To showcase the fully integrated capabilities, a preemptive-scheduling use case for Grid Engine will be automated through Unisight’s Rules Engine in a SUSE Linux cluster. In addition to informing future-proofed decisions designed to ensure optimal utilization as well as capacity planning, dynamic rules introduce an entirely new realm of action-oriented possibilities.
2. www.univa.com
❑ Metrics for monitoring and reporting upon workloads
❑ Rule-based alerts
❑ Highly dynamic use cases
❑ Just-released capabilities
❑ Q&A
2
Agenda
3. www.univa.com
3
Unisight 4.0
▪ Completely refactored implementation
▪ Monitors and reports upon Univa Grid Engine workloads
▪ Integrates with Univa Grid Engine add-ons
▪ Included with an existing license for Univa Grid Engine
4. www.univa.com
4
Unisight 4.1
▪ Compare multiple values from an attribute in an object
▪ e.g., compare ‘running’ and ‘pending’ jobs on the same graph
▪ Collect metrics automatically for NVIDIA GPUs
▪ Makes use of the Univa Grid Engine custom load sensor
▪ Create tables and charts for Docker-enabled hosts and recently
used Docker images
▪ Requires use of Docker-enabled Univa Grid Engine
▪ Import legacy reporting or accounting files
▪ Requires Univa Grid Engine 8.2.x or more recent
▪ Automatically detect and use added/removed complex entries
▪ Based on interaction with one or more Univa Grid Engine clusters
▪ (Site-defined) complexes can be used in Unisight filters, tables & charts
5. www.univa.com
5
Unisight 4.2
▪ Built-in dashboard charts for cluster and job efficiency
▪ For each job: CPU Time / ( Wallclock Time * Slots )
o May be normalized
▪ Job details including pending reasons
▪ ~ “qstat –j <JOBID>” + “qalter –w p <JOBID>”
▪ Additional types of charts
▪ e.g., bar
▪ Alerting rules – for admins only in this release
▪ New screen added to support this capability
▪ Any data available via Unisight can be used in these rules
▪ Actions include the ability to alert the admin via the Unisight UI, email
or Slack
6. www.univa.com
6
Unisight 4.3
▪ Dynamic rules
▪ Data known to Unisight can be framed as a variable
▪ Rule handlers includes a chaining capability
▪ Concurrently pending jobs as a function of time
▪ Concurrent license utilization as a function of time
▪ CPU efficiency as a function of time
▪ Memory utilization displayed in different units
16. 16
Unisight Prerequisites
▪ Supported operating systems
▪ 64-bit SUSE Linux Enterprise Server 12 and 11 SP4
▪ Network configuration
▪ MongoDB, port: 57017
▪ UniSight REST API, port: 8080, 8443
▪ Web GUI, port: 3003
Ports can be changed from the default values
▪ Hardware
▪ Each component can be installed on different machines
▪ For a single machine install
o Quad core system with 8 GB RAM and 100 GB disk
www.univa.com