Knowledge Base - Glossary

• DMF
  Data Migration Facility; the SGI Data Migration Facility (DMF) is a virtual file manager at NAS that migrates data from the Lou mass storage systems to the Spectra Logic archival tape storage systems (disk to tape) and recalls data (tape to disk) as needed. A list of DMF commands for controlling file transfers at NAS is available in KB article 250. 
• front-end node
  A node of a system that supports login and execution of tasks related to job preparation, such as compiling and file copying (as opposed to back-end node). For the different front-end nodes at NAS, see PFEs, LFEs, and mfe1.
• PFEs, pfe
  Pleiades Front-End nodes which provide a load balanced front-end environment for quick turnaround while performing file editing, file transferring, compiling, short debugging/testing sessions, and batch job submissions via PBS to the compute nodes. See KB article 181.
• endian
  Most commonly refers to the way bytes are ordered within a data item. Big-endian machines store the most significant byte first (at the lowest byte address), while little-endian machines store the least significant byte first. At NAS, Intel Fortran expects numerical data to be in little-endian order. See KB article 206. 
• rank
  A variable commonly used in MPI functions. Assigns each process a unique integer rank from 0 to (the number of processes - 1), which determines the rank of the calling process within the communicator. The master process has the rank of 0.
• OpenMP
  An API that supports multi-platform shared-memory multiprocessing programming in C, C++, and Fortran on most processor architectures and operating systems. It consists of a set of compiler directives, library routines, and environment variables that influence runtime behavior. See KB article 209.
• thread
  The smallest sequence of programmed instructions that can be managed independently by an operating system scheduler; a light-weight process. For the most part, multiple threads can exist within the same process and share resources such as memory, while different processes do not share these resources. On a multiprocessor or multi-core systems, such as those at NAS, threads can be truly concurrent, with every processor or core executing a separate thread simultaneously.
• Hyper-Threading
  Intel's proprietary simultaneous multithreading implementation used to improve parallelization of computations (doing multiple tasks at once) performed on microprocessors. For each core that is physically present, the operating system addresses two virtual or logical cores, and shares the workload between them when possible in order to decrease the number of dependent instructions on the pipeline.
• Turbo Boost
  Technology on Intel processors which automatically allows processor cores to run faster than the base operating frequency for short durations in order to maximize performance if they're operating below power, current, and temperature specification limits. The maximum frequency for Turbo Boost is dependent on the number of active cores, and the amount of time the processor spends in the Turbo Boost state depends on the workload and the operating environment. Each processor type at NAS has its own Turbo Boost requirements.
• inode
  A data structure found in many Unix and Linux filesystems; each stores all information about a filesystem object (file ownership, access mode, type of file, etc.), except data content and file name.
• rack
  Within NAS, the term "rack" refers to the physical housing that connects each of the components of the supercomputers. NAS systems are SGI Altix 4000/ICE and SGI UV 2000 architecture types. See IRU.
• LFEs
  Lou Front-End nodes; the designation of the two hosts, lfe1 and lfe2, of the Lou2 archive system cluster; operate similarly to the PFEs. See KB article 371.
• DMZ
  NAS Demilitarized Zone; file transfer servers, dmzfs1 and dmzfs2, designed to help facilitate file transfers into and out of the NAS enclave. All Lou users have an account on the DMZ servers. See KB article 146. 
• shared memory
  A structure of multiprocessor computers that allows memory to be simultaneously accessed by multiple processors. At NAS, the Endeavour supercomputer is a global shared memory system that allows a single processor to access the memory of the processors around it, if they are needed to execute its job. See KB article 410 for more information on the Endeavour system.
• SUP
  Secure Unattended Proxy; a client that allows users to perform remote operations on specific hosts within the NAS systems without the use of SecurID authentication at the time of operation. Users must obtain "SUP keys" using SecurID authentication, after which those keys can be used to perform operations from unattended jobs and/or scripts for one week after the time the key is generated. See KB article 145 for more information on using SUP.
• InfiniBand (IB)
  A switched fabric communications link used in high-performance computing that features high throughput, low latency, quality of service, and failover, and is designed to be scalable. NAS systems use InfiniBand fabric connections between nodes of the supercomputers, as well as from the supercomputers to the mass storage systems. There are two independent InfiniBand fabrics at NAS, referred to as ib0 and ib1. See KB article 237.
• IRU
  Individual Rack Unit; feature of the SGI Altix 4000 and ICE line of system architecture, the IRU is an enclosure that contains the necessary components to support the blades (containing processors and memory chips) such as power supplies, a router board for every five blades, and an L1 controller.
• NUMA
  Non-Uniform Memory Access; a computer memory design used in multiprocessing where the memory access time depends on the memory location relative to a processor, since a processor can access its own memory faster than non-local memory from another processor or memory shared between processors. See KB article 259.
• RDMA
  Remote Direct Memory Access; direct memory access from the memory of one computer into that of another without involving either one's operating system. This permits high-throughput, low latency networking, which is especially useful in massively parallel computer clusters. The InfiniBand fabrics (ib0 and ib1) that interconnect the NAS systems allows for RDMA communications via a message passing interface (MPI). See KB article 100.
• SBU
  Standard Billing Unit; a common rate used for allocating and tracking computing time at the NAS facility and the NASA Center for Climate Simulation (NCCS), using the "Westmere" nodes of Pleiades as a baseline. A job that occupies one Westmere node for one hour is charged one SBU, regardless of how many cores in the node are actually used. Use of other node types is charged based on how efficiently they are in comparison to the Westmere nodes. See KB article 271.
• SFEs
  Secure Front Ends; the bastion hosts protecting the NAS high-end computing enclave. Users must connect to the SFEs from their localhost to access the supercomputing and archival resources at NAS. See KB article 54.
• GID
  The project Group ID that is assigned to each approved project using NAS systems. A specific number of computing resources are allocated to a GID (across all users under that GID).
• GNU
  A complete, free software system, upward-compatible with Unix. Several GNU tools and libraries are available on NAS system, including the GNU Parallel shell tool, the GNU Privacy Guard (GPG) encryption system, the GNU Compiler Collection (GCC), and the GNU Debugger (GDB). 
• MAU
  Minimum Allocatable Unit; the smallest unit of hardware resource that PBS will allocate to a job. On Pleiades, one MAU equals one node. On Endeavour, one MAU equals eight cores. See KB article 271. 
• MDS
  Metadata Server; service nodes of the Lustre filesystems at NAS that manage all metadata operations such as assigning and tracking the names and storage locations of directories and files on the Object Storage Target (OST). See KB article 224. 
• MDT
  Metadata Target; a storage device on the Lustre filesystems at NAS where metadata (name, ownership, permissions, and file type) are stored. See KB article 224. 
• OSS
  Object Storage Server; service nodes of the Lustre filesystems at NAS that run the Lustre software stack, provide the actual I/O service and network request handling for the Object Storage Target (OST), and coordinate file locking with the MDS. Each OSS can serve up to around 15 OSTs. See KB articles 224. 
• OST
  Object Storage Target; storage devices on the Lustre filesystems at NAS where the data in user files are stored. See KB article 224.