The author, Wolfgang Gentzsch, was the former CEO and President of Genias Software, which originally developed the Grid Engine software that was acquired by Sun in 2000.
Today’s announcement that Univa will provide continuity and ongoing support for Grid Engine, doesn’t come as a real surprise to me. Over the past three years, Univa has invested significant time and energy in Grid Engine and made it part of their professional software and services offerings for data center optimization.
This move is clearly a win-win-win situation: Grid Engine’s thousands of users now have access to enterprise-class support at Univa, and Univa customers benefit from the extended capabilities now available for fully optimizing data center efficiency; and the Grid Engine open source community can rest assured that robust development will continue thanks to Univa’s commitment to the project.
Given this new situation, in the following, let me elaborate a bit more on the history of Grid Engine, its main functionality and characteristics, key features and capabilities, its relevance for clouds and for Univa’s data center optimization strategy.
Grid Engine Genesis
Today, Grid Engine is one of the most widely adopted schedulers in existence, utilized by thousands of users and hundreds of Fortune 500 companies around the globe. It is presently available as commercial version from Univa as well as Oracle and as source code from two sites: The “Open Grid Scheduler (OGS)” project and the “Son of Grid Engine” project. From 2001 until the end of 2010 and the closure of the original project site by Oracle, the Grid Engine open source project. Grid Engine has gained an enormous adoption within that time frame.
The Grid Engine technology came to Oracle through the acquisition of Sun Microsystems in 2010 and previously to Sun through the acquisition of Gridware Inc by Sun in 2000. Prior to that, the technology was available in two variations, CODINE (COmputing in DIstributed Networked Environments) and GRD (Global Resource Director). Throughout the nineties the technology got expanded by collaborations between Genias Software GmbH, Chord Systems, Raytheon and Instrumental Inc. The origins of the software date back to 1992 when Genias Software GmbH acquired the right to commercialize DQS (Distributed Queuing System) from Florida State University.
Relevance to Cloud
Grid Engine is a key component in many cloud infrastructures. It is used for workload orchestration, resource management and priority control by cloud infrastructure providers as well as customers operating private and hybrid clouds. With the addition of Grid Engine, Univa is now capable of offering a full data center optimization stack with private, public or hybrid cloud options to increase IT operational efficiency. In combination with the Univa UniCloud product, Grid Engine can be deployed automatically and re-sized on demand. UniCloud also provides the integration of virtual machine management with Grid Engine, so that virtual hosts under Grid Engine control can be checkpointed and migrated as the need arises, e.g. to free up key resources for top priority tasks.
Positioning in the Market
Grid Engine is one of the most widely used resource management systems, in a broad spectrum of markets such as bio-technology, electronic design automation, energy, science and industrial manufacturing. Recently, however, with Oracle’s acquisition of Sun, users and customers became somewhat irritated about Grid Engine’s future. Now, with the move of Grid Engine to Univa and its focus on the product and on corresponding services plus its synergy with other Univa products, users (and potential customers) have a solid and reliable basis for their further planning, and an exciting continuation of development can be expected.
With Grid Engine I am convinced that Univa – more than ever – will continue to solve real ‘burning’ customer needs and stay away from hypeful marketing, as proven for example with its recent Data Center Optimization Strategy announcement. In many markets, Clouds are just one dimension of a company’s data center optimization problem, and the wise approach is evolution, not revolution; optimization, not replacement.
The Community
The reaction on this move I see in our active community is all positive, e.g. from the project owners of the Open Grid Scheduler (OGS) project, from BioTeam, and others. It should also be in Univa’s interest to continue to maintain this community and the technology, because of the high potential of future customers from this user base. On the other hand, it is very clear to me, that without such a move and Univa’s commitment to Grid Engine, this great technology, team, and community would have fallen apart very soon.
The New Business Model
The current market for distributed resource management (DRM) systems and services is at least $250M, and at least half of the market is not yet unlocked. Many organizations still tamper with their own home-made solutions and thus waste valuable time, energy, and money. This market offers ample opportunities for leaders like Univa, Platform, and Altair, and a few other competitors. Add to that the Cloud and IT Infrastructure management markets, which Univa addresses with its data center optimization strategy, including its management components UniCluster and UniCloud. Now, with Grid Engine, Univa’s business model race car is complete, it has got its engine.
Grid Engine Functionality and Characteristics
Grid Engine is an industry-leading distributed resource management (DRM) system used by hundreds of companies worldwide to build large compute cluster infrastructures for processing massive volumes of workload. A highly scalable and reliable DRM system, Grid Engine enables companies to produce higher-quality products, reduce time to market, and streamline and simplify the computing environment.
Grid Engine makes it easy to create a cluster of thousands of machines by harnessing the combined computing power of desktops and servers in a simple, easy-to-administer environment. Scheduling policies can be applied to all work submitted to the cluster, ensuring high-priority jobs are completed on time while simultaneously maintaining maximum utilization of all cluster machines. With Grid Engine, any resource or software license can be monitored, tracked and scheduled to ensure applications are automatically matched to the appropriate licenses and machines in the cluster. Some of the key features and capabilities of Grid Engine are:
• Priority and Utilization Policies: Grid Engine delivers multiple scheduling policies for matching workload in the cluster to business and organizational objectives such as maximizing utilization across all machines, reducing turnaround time for jobs in the cluster, or prioritizing workload according to group, department or company affiliation.
• Dynamic Resource Management: Grid Engine continuously collects metrics from all cluster nodes, then uses scheduling strategies configured by the administrator to evaluate all pending workload and match specific job requirements to available resources.
• Resource Capacity Management: All resources in a data center or cluster are finite and must be effectively shared. Grid Engine provides powerful and flexible resource capacity control for all shared resource, including applications and licenses. Resource Quota Sets further refine the level of control for sharing resources among users, groups and departments.
• Scalability: Grid Engine can scale to a cluster of 64,000 cores (or more) in a single managed environment. A single Grid Engine cluster can contain more than 4,000 nodes, and additional Grid Engine clusters can also be added to extend scalability to 10,000 or more nodes in a multi-cluster environment.
• Performance: With Grid Engine, performance scales seamlessly and efficiently as the cluster grows in size. Grid Engine can accept over 100 jobs per second, and using the DRMAA interface a 1,000-job/second submission rate can be achieved. The Grid Engine scheduler can be configured as the cluster grows in size and number of jobs to ensure minimal time is spent scheduling and dispatching jobs to individual cluster hosts. Performance optimization is built into every aspect of Grid Engine, from fast system startup times even with multiple 100K jobs in the system to dispatching and running extremely large-scale parallel jobs across 64K cores.
• Nonstop Cluster Reconfiguration: Production clusters cannot afford to stop or suspend work for cluster modification. With Grid Engine all cluster configurations and parameters can be modified and machines can be added or removed while Grid Engine continues to run.
• Flexible Workload Types: Grid Engine supports many different workload types so a wide variety of jobs can be scheduled and run efficiently in the cluster. Most applications can run in a sequential fashion; however, some specialized applications may require either a parametric workload type for simple parallel applications or full parallel workload type for applications that need message passing. Interactive jobs can be submitted to Grid Engine providing a remote shell to the user while at the same time enforcing Grid Engine scheduling and policy control.
• Multi-Core Processor Binding: With the advent of NUMA, multi-core machines scheduling effectively to a single processor requires binding the job to specific processors or cores on the machine. Grid Engine provides superior flexibility in assigning jobs to specific cores, improving the overall utilization and runtime for most jobs.
• Security: Grid Engine can be configured to encrypt all data being transferred among Grid Engine components running on the cluster nodes. Using this secure mode, Grid Engine can also verify user certificates and ensure that components have not been tampered with by potential attackers.
• Advance Reservation: Reservations can be made to reserve Grid Engine cluster resources (such as CPU, memory, job slots or applications) for a future time and date. This feature is invaluable for cluster administrators and power users to reserve portions of the cluster for maintenance, parallel jobs or very high priority workload.
• Hybrid Cloud Enablement: Grid Engine is fully integrated with UniCloud, providing functionality to seamlessly extend a Grid Engine cluster to an external cloud computing service such as Amazon EC2 or Rackspace Cloud. It is fast and easy to create a Grid Engine cluster that runs in the public cloud or to establish a secure hybrid cloud that contains machines from your environment and the public cloud provider.
About the Author
The author Wolfgang Gentzsch was the former CEO and President of Genias Software which originally developed the Grid Engine software acquired in 2000 by Sun. Wolfgang left Sun in 2004, to lead the North Carolina Grid and Data Center, and was the Chairman of the German D-Grid and a member of the Board of the Open Grid Forum. He is currently an Advisor to the EU Distributed European Infrastructure for Supercomputing Applications, DEISA.