Moabcon 2012 Technical Sessions

Presenter: Josh Butikofer

Abstract:
Moab has a lot of “policies.” But what exactly constitutes a policy? How are they best configured? Comes learn about some of Moab’s most used policies and how to get them working in your environment.

Presenter: Dan Croft

Abstract:
A brief introduction to the world of cloud from an HPC point of view. We will explore the core concepts of a public and private cloud setup and see how these can benefit traditional HPC workloads. We will look at Moab’s new cloud model and how it can be used in a traditional HPC environment and look ahead to where cloud and HPC are going in the future.

Customer Case Study 1

Presenter: Jenett Tillotson, Indiana University

Abstract:

• An overview of Indiana University cluster compute systems and user base
• TORQUE queue design
• Compare and contrast LoadLeveler and TORQUE
• Priority algorithms focused on fair share, X-factor, and parallel job prioritization
• User limits to prevent queue stuffing
• Job prioritization using the QOS system versus node reservations

Customer Case Study 2

Presenter: Michael Jennings, Lawrence Berkeley National Laboratory

Abstract:
TORQUE has long offered the ability to periodically monitor compute node health and report misconfigured or failed nodes. Sites often overlook this feature or use ad hoc, homegrown scripts to perform the checks they need. Mr. Jennings will present a flexible, extensible alternative, which are both a working implementation and a model/framework for node health checks.

Presenter: Luis Silva and Ken Baldwin

Abstract:
In this presentation we will be covering the new exciting features of Moab Cloud Suite 7.0 xCAT edition. For this release, we focused heavily around enhancing the user experience and providing our customers with tools that make their lives easier.One example of this is our Service Catalog Manager, where our customers can define service dependency flows to more easily automate the provisioning process for their users. With it, administrators can easily define new catalog entries and so that their users can submit new service requests without having to go through tedious and costly bureaucratic processes. We will also be doing a demo focused around “a day in the life” of an admin and an end user, showing how admins can create cost centers in the Moab Accounting Manager, allocate users to them, creating new service templates and having users request these in just a few minutes.

Case Study 1

Presenter: Troy Baer, Senior HPC System Administrator, NICS

Abstract:
Techniques for Balancing Capability and Capacity Workloads for Maximum UtilizationWorkloads on HPC resources are generally considered either “capability” (a few large jobs) or “capacity” (many small jobs). However, it is possible to have both types of workloads coexist on the same system using a few straightforward scheduling techniques. This talk will discuss how those techniques have been applied at the National Institute for Computational Sciences.

Case Study 2

Presenter: Frederick Lefebvre, Compute Canada

Abstract:
Colosse is a mid-size cluster used for academic research deployed at CLUMEQ/Universite Laval in late 2009. This 7680 cores system, initially using Sun Grid Engine, was transitioned to TORQUE/Moab in 2012. This talk will focus on our experience in migrating from Grid Engine to Moab.

Presenter: Nathan Wells

Abstract:
Get a close up look at Moab Cloud Suite – CSA Edition with a demonstration, and presentation on how the integration works.

Partner Presentation

Presenters: Mak Ghangurde, James Matthews

Abstract:
An overview of HP Cloud Service Automation (with Moab Cloud Suite) – a new partnership and product offering between HP and Adaptive Computing that offers enterprises the foundation to venture into the cloud and evolve to the cloud needs of tomorrow. This session will focus on CSA’s new Advanced Workload Management functionalities that can help enterprises dynamically allocate resources and adapt to changing resource requirements.

Presenter: Brian Christiansen

Abstract:
Configuring Real-World SLA’s in Moab.This session will go over several real-world examples of SLA’s and how they can be configured in Moab. Examples include:

• Sharing heterogeneous sets of nodes that have specific owners.
• Giving node access to developers during the day for testing.
• Assigning priority based on job size.

Customer Case Study

Presenter: Matt Ismail, University of Warwick

Abstract:
UK University HPC service provision is currently expanding in both scope and scale. Universities are now expected to build on their established central HPC services to contribute resources into geographically distributed national facilities and construct regional e-infrastructure intended to catalyse engagement with industry. This talk will describe how Moab is applied at the Centre for Scientific Computing, University of Warwick to implement the diverse set of policies and service levels required of a multi-tier HPC facility. There will be particular emphasis on the fundamentals of scheduling such as ensuring fairness, delivering quality of service and managing mixed workloads from various application domains and external organisations.

Presenter: Scott Bateman

Abstract:
Sharing resources between 2 compute clusters can grant users from both the benefit of on-demand elasticity in their resource pool. Moab’s HPC Cloud Bursting offering connects separate groups of compute resources and allows one to assume control over nodes from the other based on sophisticated, customizable rules of priority and preemption.

Presenter: Scott Jackson

Abstract:
This session will introduce the Moab Accounting Manager and describe how it can be used for usage tracking, charge accounting and allocation enforcement. We will demonstrate using Moab Accounting Manager in an HPC context to establish per-project limits on the use of compute resources.

Customer Case Study

Presenter: Andrew Caird, University of Michigan

Abstract:
Using Gold, Moab, and Torque to Run a Campus-wide Allocation-based HPC Service

The University of Michigan has moved some condo-style cluster workload to an allocation-based system that is based on Gold with some extensions, the Moab Identity Manager, Moab, and Torque. Using those tools we support over 150 research projects and maintain high-resource utilization in a financially sustainable hardware and software ecosystem.

Presenter: Nick Ihli

Abstract:
With Moab, NICE DCV and Enginframe site’s can consolidate expensive high-end GPU workstations providing users greater access to high-performance visualization of complex 3D interactive application alongside traditional HPC workload. The rendering is accomplished in the “cloud” and instead of moving data, pixels are sent to the remote display.

Presenter: Douglas Wightman

Abstract:
Moab is the most advanced scheduler on the planet. This session will explain how to take advantage of Moab’s patented reservation scheduling and preemption engine. Learn how to maximize your system’s resources while delivering the fastest response to your users.

Presenter: Scott Jackson

Abstract:
This session will describe the use of the Moab Accounting Manager within a cloud context and will go into greater depth about customizing the usage record and creating a customized departmental hierarchy.

Presenter: Ken Nielson

Abstract:
How Does TORQUE 4.0 Change Your JobTORQUE 4.0 added new capabilities, which in practice will change the way administrators need to tune their systems. This presentation will discuss key new features and how they change the behavior of TORQUE as well as how to tune the new features to take advantage of the full power of TORQUE. Features to be discussed include setting up a MOM hierarchy; at what number of nodes do you need to use a MOM hierarchy, setting a thread pool size, TCP communications change and job radix.

Presenter: Shawn Hoopes

Abstract:
Have you ever wondered why your jobs end up in a Deferred state? In this session, you will learn why and much more, by understanding the Moab scheduling cycle and determine where to troubleshoot job issues along this path. You will learn how to effectively use the mdiag and other troubleshooting commands, how to look at log and event files and check the messages, notify Admins when handling certain events, how to track system failures, and learn how to use the included diagnostic scripts.

Presenter: Ryan Chabot

Abstract:
Upgrading Moab HPC Suite can be complicated at times and it is vital to understand the process of upgrading between major versions. This session will cover the installation / upgrade best practices as presented by members of the Adaptive Computing support group. Topics include: Operating System configuration, verifying proper permissions, ensuring correct file locations, installing updated binaries, database upgrades and insights in to what and upgrade modifies and installs. We will also discuss some of the most common trouble areas related to upgrading and how to troubleshoot and fix these issues. After this course, you should be able to migrate or upgrade your Moab HPC Suite with a full understanding of the process.