h3. HPC Diskless Cluster on Sun Blade 6000
Sun's recent deployment of the Sun Constellation System at TACC shows how serious we are at HPC, and how the blade systems fits nicely in this environment. This Blade 6000 Linux diskless cluster based on x64 AMD Opteron systems is set to run HPC applications for academic research and training, in the Meteorology and Geophysics areas, running atmospheric simulation and numerical predictions models, with MPI libs and visualized at powerful Ultra 20 M2 workstations.
A major south american University has just received a Sun Blade 6000 system based cluster, under the Sun Academic Initiative, which shows how Sun is committed both to HPC and the Academia. Sun has HPC solutions that scales from small department clusters up to the most powerful computers in the world, like the [TACC Ranger|http://blogs.sun.com/marchamilton/entry/more_ranger_facts_and_figures], from as small as 10 nodes to as big as 60,000 cores (like the Ranger setup).
The Blade 6000 will run a cluster for testbed and research on parallel scientific codes for the related research areas, providing researchers and students with more contact with HPC systems, and creating a place for training and developing expertise in parallel programming, code optimization, and model researching and tweaking.
*{+}Setup{+}*
The system is implemented in a Blade 6000 chassis populated with Sun X6220 Server modules with 2 sockets AMD Dual-corel Opteron processors, and Sun Ultra 20 M2 workstations, for visualization and programming. The Ultra 20 M2 workstations are running Solaris 10, while the Blade cluster is running Linux. The system is networked using standard Gigabit Ethernet.
The cluster is diskless, meaning that only the first server module has actually an installed operating system, which provides boot services and NFS root exports for the diskless systems. All diskless nodes share the same (customized for concurrent access) root filesystem, allowing for:
* Easy software maintenance, less administration overhead (install in one node, and all nodes can use it)
* Easy to upgrade the cluster system (just change to another chroot and kernel)
* Flexible setup (could switch to a Solaris Cluster mode in a matter of minutes)
The cluster is running a MPI based model for atmospheric simulation, the BRAMS model and GrADS (Grid Analysis and Display Software) post processing / visualization tool. The BRAMS/RAMS model is a multipurpose, numerical prediction model designed to simulate atmospheric circulations spanning in scale from hemispheric scales down to large eddy simulations (LES) of the planetary boundary layer, tailored for the tropics region of Brazil. The Grid Analysis and Display System (GrADS) is an interactive desktop tool that is used for easy access, manipulation, and visualization of earth science data.
*{+}Benefits{+}*
* Compact and dense (but still cool) for deployments with space or thermal constraints
* Easy to assemble and to service, less cables, less fans and power supplies, increased reliability
* Scalability and flexibility with NEMs and EMs for several IO's (GbE, 10GE, Infiniband, fiber, SAS)
* Eco-friendly with less power supplies and intelligent load-balancing PSUs
* No compromise on performance when going with the blade solution: powerful systems as much as a "normal" servers (no "subpar servers")
* Open Architecture, based on industry standards technology (PCI Express, infiniband, SNMP, IPMI), and pick your favourite architecture: SPARC, AMD or Intel blades
*{+}Recommended Configurations{+}*
Sample configuration for a small (to medium) academic cluster to run Meteorology simulation models.
||System||Configuration||
|Blade Chassis| * Sun Blade 6000
* Sun Blade 6048|
| Compute nodes | * Sun X6220 Server Module
* 2-socket Opteron dual-core 2.0GHz
* 4GB RAM
* diskless |
|Interconnection| * Gigabit Ethernet via NEM
* 10GE Express Modules
* Infiniband Express Modules|
| Workstations | * Ultra 20 M2
* dual-core Opteron 2.6GHz
* 4GB RAM
* NVIDIA Quadro FX1500 video cards|
| Operating Systems | * Linux
* Solaris 10|
|Compilers| * Intel Fortran and C/C++ Compilers and Math Libraries
* GCC/Gfortran
* Sun Studio|
|Parallel Libraries| * MPICH
* OpenMPI, SunCluster Tools in Solaris
* MVAPICH|
|Applications| * BRAMS atmospheric model
* GrADS postprocessing tool
* Sun Grid Engine for job management
* Ganglia Monitoring System
* Sun xVM Ops Center|
*{+}Resources{+}*
* [Sun HPC Community Portal|http://hpc.sun.com]
* [Sun HPC Solutions|http://www.sun.com/hpc]
* [Sun Blade 6000/6048 Open Modular Architecture (pdf)|http://www.sun.com/servers/blades/6000/arch-wp.pdf]
* [Petascale Computing with Sun (pdf)|http://www.sun.com/servers/hpc/docs/pathways_to_petascale.pdf]
* [BRAMS Model|http://www.cptec.inpe.br/brams/]
* [GrADS tool|http://grads.iges.org/grads/]
[[+Back to Sun Blade Modular Solutions+|http://wikis.sun.com/display/BladeSystems/Sun+Blade+Modular+Solutions]]
Sun's recent deployment of the Sun Constellation System at TACC shows how serious we are at HPC, and how the blade systems fits nicely in this environment. This Blade 6000 Linux diskless cluster based on x64 AMD Opteron systems is set to run HPC applications for academic research and training, in the Meteorology and Geophysics areas, running atmospheric simulation and numerical predictions models, with MPI libs and visualized at powerful Ultra 20 M2 workstations.
A major south american University has just received a Sun Blade 6000 system based cluster, under the Sun Academic Initiative, which shows how Sun is committed both to HPC and the Academia. Sun has HPC solutions that scales from small department clusters up to the most powerful computers in the world, like the [TACC Ranger|http://blogs.sun.com/marchamilton/entry/more_ranger_facts_and_figures], from as small as 10 nodes to as big as 60,000 cores (like the Ranger setup).
The Blade 6000 will run a cluster for testbed and research on parallel scientific codes for the related research areas, providing researchers and students with more contact with HPC systems, and creating a place for training and developing expertise in parallel programming, code optimization, and model researching and tweaking.
*{+}Setup{+}*
The system is implemented in a Blade 6000 chassis populated with Sun X6220 Server modules with 2 sockets AMD Dual-corel Opteron processors, and Sun Ultra 20 M2 workstations, for visualization and programming. The Ultra 20 M2 workstations are running Solaris 10, while the Blade cluster is running Linux. The system is networked using standard Gigabit Ethernet.
The cluster is diskless, meaning that only the first server module has actually an installed operating system, which provides boot services and NFS root exports for the diskless systems. All diskless nodes share the same (customized for concurrent access) root filesystem, allowing for:
* Easy software maintenance, less administration overhead (install in one node, and all nodes can use it)
* Easy to upgrade the cluster system (just change to another chroot and kernel)
* Flexible setup (could switch to a Solaris Cluster mode in a matter of minutes)
The cluster is running a MPI based model for atmospheric simulation, the BRAMS model and GrADS (Grid Analysis and Display Software) post processing / visualization tool. The BRAMS/RAMS model is a multipurpose, numerical prediction model designed to simulate atmospheric circulations spanning in scale from hemispheric scales down to large eddy simulations (LES) of the planetary boundary layer, tailored for the tropics region of Brazil. The Grid Analysis and Display System (GrADS) is an interactive desktop tool that is used for easy access, manipulation, and visualization of earth science data.
*{+}Benefits{+}*
* Compact and dense (but still cool) for deployments with space or thermal constraints
* Easy to assemble and to service, less cables, less fans and power supplies, increased reliability
* Scalability and flexibility with NEMs and EMs for several IO's (GbE, 10GE, Infiniband, fiber, SAS)
* Eco-friendly with less power supplies and intelligent load-balancing PSUs
* No compromise on performance when going with the blade solution: powerful systems as much as a "normal" servers (no "subpar servers")
* Open Architecture, based on industry standards technology (PCI Express, infiniband, SNMP, IPMI), and pick your favourite architecture: SPARC, AMD or Intel blades
*{+}Recommended Configurations{+}*
Sample configuration for a small (to medium) academic cluster to run Meteorology simulation models.
||System||Configuration||
|Blade Chassis| * Sun Blade 6000
* Sun Blade 6048|
| Compute nodes | * Sun X6220 Server Module
* 2-socket Opteron dual-core 2.0GHz
* 4GB RAM
* diskless |
|Interconnection| * Gigabit Ethernet via NEM
* 10GE Express Modules
* Infiniband Express Modules|
| Workstations | * Ultra 20 M2
* dual-core Opteron 2.6GHz
* 4GB RAM
* NVIDIA Quadro FX1500 video cards|
| Operating Systems | * Linux
* Solaris 10|
|Compilers| * Intel Fortran and C/C++ Compilers and Math Libraries
* GCC/Gfortran
* Sun Studio|
|Parallel Libraries| * MPICH
* OpenMPI, SunCluster Tools in Solaris
* MVAPICH|
|Applications| * BRAMS atmospheric model
* GrADS postprocessing tool
* Sun Grid Engine for job management
* Ganglia Monitoring System
* Sun xVM Ops Center|
*{+}Resources{+}*
* [Sun HPC Community Portal|http://hpc.sun.com]
* [Sun HPC Solutions|http://www.sun.com/hpc]
* [Sun Blade 6000/6048 Open Modular Architecture (pdf)|http://www.sun.com/servers/blades/6000/arch-wp.pdf]
* [Petascale Computing with Sun (pdf)|http://www.sun.com/servers/hpc/docs/pathways_to_petascale.pdf]
* [BRAMS Model|http://www.cptec.inpe.br/brams/]
* [GrADS tool|http://grads.iges.org/grads/]
[[+Back to Sun Blade Modular Solutions+|http://wikis.sun.com/display/BladeSystems/Sun+Blade+Modular+Solutions]]