Streamline's software stack for high performance and complex computing
High performance computing applications have highly specific requirements when it comes to the computing environment needed to deliver the maximum benefit from licenses and other high performance computing investments. The range of possible software components, middleware, parallel environments, file systems and operating systems is vast and Streamline’s reputation rests on being to manage this fragmented landscape of open source and proprietary software into a consistent, tested, supportable and manageable software stack, specifically assembled and configured for high performance computing environments.
Streamline's parallel computing software stack
Streamline Computing integrates a number of parallel computing components into its software stack. These include OpenMPI, MPICH2, SunHPC, MVARPICH, MyrinetMX*, OpenMX, HPMPI* and the Modules environment to easily switch between different MPI environments. Generally OpenMPI is installed as the default MPI environment but other flavours of MPI can be substituted or added alongside depending on the needs of the system and user environment.
Batch scheduling is provided by the open source Sun Grid Engine (SGE) although Streamline’s software stack can also accomodate other schedulers such as OpenPBS, Torque and Maui. The default SGE configuration includes serial and parallel queues and additional queues can be configured during the build process.
Streamline’s HPC software stack also includes installations of the most popular compilers and scientific libraries. While some are open source, some require commercial licenses. Installed packages include the C, C++, F77 and F90 GNU compilers as well as base installations for commercial Intel* and PGI* compilers. Installed scientific libraries include FFTW, ScalarPack, Lapack/Blas (Intel MKL*, ACML, Goto and Atlas). Each Streamline system also comes with evaluation installations of Allinea Software’s parallel debugger DDT and also OPT, their parallel optimizer.
The software stack developed and supported by Streamline Computing also includes the latest CUDA programming and run-time environments for GP-GPU enabled systems.
The Gluster parallel file system is the preferred file system on all Streamline clusters although NFS file systems can also be provided where necessary.
Supportable Linux operating systems
Streamline Computing’s preferred operating system distribution is Scientific Linux; a Linux distribution produced by the Fermi National Accelerator Laboratory and CERN (the European Organization for Nuclear Research). It is a freely available operating system based on RedHat Enterprise Linux and offers the benefit of long term support while at the same time being able to take advantage of Enterprise-class deployment. For industrial environments and to support certain ISV codes, Streamline also offers RedHat Enterprise Linux* as a supported operating system on its clusters.
Streamline Computing offers traditional clusters where the operating system is deployed to hard disks resident on compute nodes. However, where practical, Streamline can also offer disk-free systems where the compute node operating system can run in memory as a ramdisk, thereby reducing running costs and the expense of maintain- ing excessive numbers of hard disk drives.
High bandwidth, low latency networks
Streamline Computing has been at the forefront of high performance networks on clusters since its creation. Streamline has continued to build this expertise and can offer Myrinet, Infiniband and 10G networks on all its systems. Streamline’s software stack includes support for the OFED pack- age, QLogic’s InfiniPath drivers, OpenMX and MyrinetMX* - all of which can be added and supported as optional installations.
Streamline can provide Myrinet-10G, Infiniband and 10G networks on all systems to suit all customer’s needs.
* commercial licensing may be required