Univa Grid Engine

Performance Begins with Univa Grid Engine

Performance of modern workload management systems can be defined many ways depending on the primary use and dependency on the system. For many Grid Engine users downtime is most unacceptable. Univa agrees and that's why we view stability as the foundation of performance that enables the workload management system to maximize throughput. The stability found in Univa Grid Engine has allowed the design of unique features that extends performance to make Grid Engine scream in the most demanding data centers.

Unique Performance Features

• Best Support

  As a customer, you don't just get support when you need it from Univa. You gain access to our exclusive expertise in the scheduler, policies and best practices. Our customers regularly learn from our knowledge and insight and apply it to their unique environment and configuration. Often this leads to customer requirements on our roadmap.

• Job Classes

  Increase your operational efficiency with Univa's advanced job management via Job Classes – the single largest functional improvement to Grid Engine in five years. Job Classes – essentially job or application templates - allows you to define defaults or group your workloads and streamline their management from attribute control and policy implementation to accounting.

  Benefits
  • Tuned workload means significant improvements in throughput
  • Avoid rogue jobs and poor cluster utilization by badly tuned workloads

• Short Jobs

  Grid Engine users with an extreme number of short jobs can see dispatch times explode when they stuff hundreds of thousands of jobs into the scheduler. Univa's support for Postgres database job spooling balances speed of submission with reliability in high volume clusters with lots of small jobs. We know this method works – it's production proven by some of our most demanding customers.

  Benefits
  • Dramatic improvements in throughput at scale

• Core & Non-Uniform Memory Access (Numa) Binding

  We implemented changes in the scheduler to optimize the use of modern computer architectures and topology characteristics such as sockets, cores and memory banks. Our design ensures the best repeatable performance across different server vendor designs through automated, optimized core and NUMA selection. We can also guarantee that specific jobs have exclusive access to the required cores and memory segments for optimal performance.

  Benefits
  • Including NUMA binding improves application performance
  • Moving scheduling to the Master improves decision making and avoids collisions by guaranteeing a binding

• Fair Urgency

  A new scheduling algorithm that helps to ensure a balanced utilization of critical resource pools such as file servers. Fair Urgency is a standard means to avoid overloading and performance degradation.

  Benefits
  • Provides equitable access to critical resources
  • Elimination of overloading maintains performance at peak

• New Documentation & Improved debugging and diagnostics

  We rewrote the documentation New to help you find administration and configuration information. These improvements and diagnostic additions help you to find the root cause of an issue without a need to reproduce the problems and avoid those issues before they hit you again and cost you effort and utilization.

  Benefits
  • Reduce the time spent on diagnosing job related issues by up to 90%
  • Saves time again and again

• Resource Maps

  When we were working on the NVIDIA GPU integration we realized that a mechanism to map resource units in use to jobs was not easy in Grid Engine. Specifically when a job requested GPUs on a GPU-enabled host there was no easy way to tell Grid Engine that a particular GPU was attached to a job – so don't use it for another job. Also there was no easy way to tell the Grid Engine Scheduler that a specific GPU should be used for a job. So we created Resource Maps an extension to consumable resources and implemented using a new Univa Grid Engine complex attribute type called RSMAP. This is a general-purpose extension.

  Benefits
  • Eliminates resource conflict

For a complete list of new functionality please read the release notes which can be found here

What does this mean for your business?

Univa is evolving Grid Engine at a rapid pace. In the past year alone we have made substantial advances in stability and the many new features have helped our customers drive down the costs of running Grid Engine. We are the one company with the expertise and means to further enhance Grid Engine, and that creates substantial value to Univa Grid Engine users.

Our global presence and extensive partner network provide you with the power to manage your entire computational space, no matter its size or where it is deployed. Additionally, our product is backed by unsurpassed expertise and unbeatable services and support. After all, we are the Grid Engine gurus who wrote the product to begin with!

Gain Insight with Analytics

» Read the data sheet

UniSight is a Reporting and Analytics tool that allows organizations to measure, track and chargeback usage on Univa Grid Engine Clusters. UniSight provides organizations with the insight they need to make better decisions

Resources can be practically anything, people, software, disk usage, etc...

• Comes pre-configured with simple reports
• Analytics drill-down allows for ad-hoc reporting
• Any metric collected by Univa Grid Engine can be reported
• Available only to Univa Grid Engine Customers

Organizations need to understand Usage on the clusters:

• Are the correct users getting access to the cluster?
• Are Grid Engine policies working?
• Can we chargeback to departments for their usage?
• Do we need to purchase more hardware?
• Do we need to purchase more licenses?
• How long do jobs wait in the queue?

UniSight has further simplified the administration overhead by removing ARCO, and this not only provides a product with unrivalled usability, but also unbeatable reliability.

Key UniSight Features and Capabilities

Univa License Orchestrator

Most Grid Engine users have long desired to manage applications as a resource with confidence and at scale. Often monitoring license usage or using custom home-grown scripts was the only solution. Until now.

Univa License Orchestrator prioritizes the sharing of limited and expensive application license features according to business objectives by incorporating availability into Univa Grid Engine scheduling decisions. Univa License Orchestrator enables maximum workload throughput for users, groups or projects with flexible sharing policies and simple configuration bringing administration efficiency.

Univa License Orchestrator is designed by the same people who developed Grid Engine – which has been battle-tested in thousands of mission critical environments across the globe – and this allows tight integration inside the scheduler which can only be found in Univa Grid Engine.

Learn More »

Share Hadoop with Univa Grid Engine

Univa software is unique because it simplifies the challenges enterprises face when adopting Big Data solutions like Hadoop.

The foundation of Big Data is infrastructure. Without the infrastructure there is no support for the applications that run on top of it. Infrastructure is hardware and software that includes industry standard servers, storage, networking and clustering software. The most important element is the software that supports the applications. This is where Univa lives.

Univa Grid Engine is a platform used to manage the sharing of high demand resources across business units, groups and priorities. Univa Grid Engine is deployed in support of all types of mission critical computing – from technical computing applications that design new safe cars or electronics we use every day; enterprise applications that coordinate fleet and crew; converged infrastructures that support both Hadoop and other applications.

Where do you host the applications?

Buy another cluster?

How do you share?

Building and managing clusters is what our customers do and its what we know. Univa Grid Engine unifies Big Compute and Big Data workload management by making it possible to fully leverage and share existing clusters – ensuring the infrastructure is future-proofed.

Benefits

As Big Data technologies enter the enterprise, they must coordinate and integrate with existing systems management best practices in the data center. While benefits from Big Data applications are immense they can be quickly undermined by poor utilization in a silo or the inability to share.

Univa offers a range of benefits to Big Data applications

• Shared infrastructure reduces the costs of deploying Hadoop by up to 50%
• Policy driven scheduling increases utilization and control
• Support for dynamic and multiple instances of Hadoop and other applications on a shared cluster
• Sharing supports high utilization and resource availability

Cloud Enable Grid Engine

UniCloud is software that plugs Univa Grid Engine into any cloud management system or service. UniCloud enables public and private cloud resources to be utilized with existing HPC workflow.

Automation tools will drive the cloud-enablement of Grid Engine clusters and UniCloud addresses dynamic provisioning and orchestration of application workloads in response to changing demand and predefined policies.

• UniCloud fully provisions applications to standalone, virtual or Cloud servers
• UniCloud can automatically scale application environments in response to increasing workload
• Within policy, UniCloud can move resources from another running application to a higher priority workload
UniCloud will delete under-utilized servers, thus continuously optimizing the environment to reduce costs

» Learn More