To specify your requirements, obtain pricing, or get more information on Linux Labs offerings click here.
Or you can email us:
Nimbus™ can be considered the best-of-class operating system for Linux-based supercomputing, combining the best features of SSI (Single System Image) clustering with commercial refinements which bring polish, support, and ease of installation into a technically superior and feature-rich user environment.
Nimbus™ is a full Linux-based operating system for HPC (High Performance Computing) clusters. Based on a foundation of Fedora, this distribution fuses the popular Linux distro with specialized software for next-generation HPC capabilities, in addition to support for high-end communications hardware using a simple installation CD set.
In order to support SSI clustering, Nimbus™ includes bproc from the Clustermatic, the HPC architecture developed at Los Alamost National Labs (http://www.clustermatic.org) to which Linux Labs is a contributor. Bproc, the foundation of Clustermatic, is a performance-oriented "from the ground up" cluster architecture which is widely adopted as the industry standard for unifying many Linux boxes to appear and to act as a single machine. Linux Labs adds features to encompass everything from low-level machine setup all the way up to programming support.
The net result includes the most robust system reporting around the most efficient system monitoring, out-of-the-box setup for high-performance backplanes, checkpointing (the ability to stop and start cluster jobs), and even the capability to cluster non-parallelized system processes. All Redhat and Clustermatic packages are streamlined to better integrate into the system as a whole, with emhpasis on simple installation and for ease of use. In addition, the system is enhanced to integrate the very powerful Maui batch scheduler to facilitate effective allocation and use of cluster resources.
System resource monitoring is accomplished using a modified Supermon (a lightweight systems monitoring system) as the back-end, coupled with Ganglia in order to provide complete system statistics with a simple graphical interface to minimize impact on system performance. Linux Labs has done extensive work on Supermon extending its functionality to include filesystem statistics without compromising it's goal of minimizing impact on system resources. This is accomplished, in part, by modifying the Linux kernel to provide filesystem information without the need to take any locks within the kernel. Another advantage of this system is that a complete sample can be taken with a single system call, rather than the dozens required to read the same information from the proc filesystem or the sysctl interface.
Benefits:
Easy to install and maintain - Base installation setup is a simple CD install procedure. ONE system image. ONE set of libraries. ONE installation onto the master (no node setup!), ONE process stack, ONE console, and ONE steering wheel!
Checkpointing - Checkpointing is a technique whereby an image of a running application is stored periodically so that it may be restarted at that same point at a later date. It is primarily useful for applications that have long run times and no built-in method to save results periodically.
Industry Standard - Bproc, started at Los Alamos National Laboratory, has emerged as the industry standard for Single System Image across a High Performance Computing cluster.
Versatile and Intuitive- Ready out-of-the-box for HPC backplanes, advanced self-discovering network topologies such as FNN (Flat Network Neighborhood), Fault-Tolerant MPI, Ganglia Web-based performance monitoring, etherconsole, plug-and-play node capability, GUI tools, and more!
Current and Powerful feature set- LinuxBIOS-ready with tools, UML (User Mode Linux) support for virtual clustering, Maui Job Scheduler, PVFS 1.6.2 support, Cryptographic File System support (CFS). Nimbus™ also works with the Intel C Compiler.
Commercial Certainty - Yes, you can call us for help setting it up. Yes, we can come to visit. Yes, we respond to email. Yes, we give classes. Yes, we can help you parallelize your applications. Yes, we can recommend the best topology for your cluster. Yes, you get the source code too. Yes, we can help you get your applications to work as well.
Security - IronPenguin offers the ability to grant fine grained capabilities to executables rather than all or nothing suid root. In addition it offers strictly enforced chroot and jails that even a root user cannot escape.
License:
- Base system is GNU GPL, add-ons are various. System is sold on a per-processor basis, support available 24x7.
Features
- Single system Image (SSI)
- Commercial support, reliability and service
- Warm swappable nodes
- Real-time monitoring tools
- Partitionable Cluster
- Single installation station
- Native User Mode Linux support
- Enhanced Ganglia monitoring
- Flat Network Neighborhood (FNN)
- 2.6 stable linux kernel
- Global shared memory available(Cray-compatible shared memory)
- 15 terabyte volume partitions
- Plug-n-play Myrinet support and optimizations
- Plug-n-play Dolphin support and optimizations
- GPL license
- Maui job scheduler
- Checkpointing
- PVFS support included
- Can cluster non-parallelized system processes
- Encrypted Filesystem Support
This document makes a comparative analysis of the various clusterware that are in the market.
Application space/ Business cases
- Linux clusters have made great strides in the field of HPC in the last few years for the simple reason that commodity off the shelf hardware is much more affordable than specialized hardware. With advances on consumer CPUs, memory, and chipsets, raw performance easily rivals purpose built supercomputers for most applications.
- In spite of all of the advantages of commodity clusters, they have historically suffered from a few drawbacks that have limited adoption. Primarily, ease of use, and ease of administration. Nimbus™' Single System Image aims to solve both problems. A Nimbus™ cluster presents a single system image to the user and administrator, making the use of the cluster much more like using a conventional supercomputer. In a conventional cluster, each compute node has a complete operating system with it's own user database, filesystems, and configuration. Even with various tools designed to automate software installation, administration is substantially more time consuming, complex, and error prone than for a single system. With user administration, configuration, software installation, and job control unified to a single image, Nimbus™ avoids these pitfalls. Downtime and errors are minimized, and administrative expenses are reduced. All of this results in a lowered Total Cost of Ownership.
- For the user, the simplified system image means more time spent on computation, and less on inventing yet another way to make their applications run on a loosely coordinated pile of PCs.
- Unlike other approaches to Single System Image clustering, Nimbus™ just works. All of the relevant software has been pre-assembled and configured so that it will work out of the box. There is no need to pull together bits and pieces from all over and learn how to configure and compile everything and then figure out how to make the pieces work together, LinuxLabs has done all of that for you, there is no need to reinvent the wheel.
