A significant part of the material of this glossary was adapted from material originally written by Gregory V. Wilson which appeared as "A Glossary of Parallel Computing Terminology" (IEEE Parallel & Distributed Technology, February 1993), and is being re-printed in the same author's "Practical Parallel Programming" (MIT Press, 1995). Several people have contributed additions to this glossary, especially Jack Dongarra, Geoffrey Fox and many of my colleagues at Edinburgh and Syracuse.
Original version is from NPAC at <URL:http://nhse.npac.syr.edu/hpccgloss/>
Original author: Ken Hawick, khawick@cs.adelaide.edu.auSee also the index of all letters and the full list of entries (very large)
Sections: A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
LAN (n.) Local area network, a network of multiple interconnected data terminals or devices within a local area to facilitate data transfer. Most notable of LAN topologies is ethernet, token ring, FDDI, etc.
LAP (n.) link access procedure is a modified form of HDLC that CCITT specified for X.25 networks. LAP-B is link access procedures- balanced and is the X.25 implementation of SDLC and similarly, LAP-D is the ISDN and frame relay implementation of SDLC.
LAPACK (n.) A linear algebra software package, which has been mounted on a wide range of platforms. It evolved from the older LINPACK package from Netlib. See also ScaLAPACK.
latency (n.) The time taken to service a request or deliver a message which is independent of the size or nature of the operation. The latency of a message passing system is the minimum time to deliver a message, even one of zero length that does not have to leave the source processor; the latency of a file system is the time required to decode and execute a null operation. See also startup cost.
lazy evaluation (n.) A Scheduling policy under which no calculation is begun until it is certain that its result is needed. This policy contrasts with the eager evaluation used in most programs, but is often used in functional and logic programming. See also dataflow, dependence, dependence graph, futures.
LD-1 (n.) An integrated T1 with possibly voice, data and frame relay circuits. Fractional digital service hierachy level 1 with service much the same as FT-1 except the service is integrated with voice, data, video and frame relay.
light-weight process (n.) A process which executes concurrently with other processes, in the same address space and in an unprotected fashion. Light-weight processes are used by systems such as MACH to reduce the overhead of process start-up.
linear speedup (n.) Speedup that is directly proportional to the number of processors used. According to Amdahl's Law, linear speedup is not possible for a problem that contains any sequential portion, no matter how small. Gustafson's Law however, states that linear speedup can be achieved if the problem size is increased as well as the number of processor employed. See also superlinear speedup.
linear vector scan (n.) a table lookup algorithm for pipelined vector processors that in a single operation compares a large number of contiguous elements of the table against the key.
link (n.) A one-to-one connection between two processors or nodes in a multicomputer. See also bus.
link loading (n.) The amount of communication traffic carried by a link, or by the most heavily loaded link in the system. As link loading increases, both latency and contention are likely to increase. See also bisection bandwidth.
linked array (n.) a data structure designed to allow the joining of various sized lists so that the inserting and deleting of the list elements can be done in parallel without contention.
linked triadic operation (n.) act of executing a pair of vector operations, such as V+S*V as if they were a single, longer operation by taking the output of the first pipelined functional unit and directing it to the second pipelined functional unit, avoiding a trip to main memory and back.
LINPACK (n.) A linear algebra software package, which has been mounted on a wide range of platforms. It has now been superceded by LAPACK. (n.) also a set of widely quoted performance benchmarks based on linear algebra and available from the National Software Exchange.
LIPS (n.) logical inferences per second; procedure calls per second in a logic programming system.
live variable(n.) A variable visible to a process and whose value can be changed by actions taken outside that process. For example, if a variable is shared by two processes, one of which can write to it at some point, then that variable is live within the other process. See also race condition, shared variables.
livelock (n.) A situation in which a process is forever blocked because another process has preferential access to a resource needed by both processes.
LIW(n.) Long Instruction Words: the use of long (64 or more bits) instruction words in a processor to improve its ability to pipeline calculations.
LMI (n.) Local Management Interface, a protocol, with 4 different versions, used to control the local interface from a routing device to the IPX Switch. Also used for configuration, flow control and maintenance of the local connection.
load balance(n.) The degree to which work is evenly distributed among available processors. A program executes most quickly when it is perfectly load balanced, that is when every processor has a share of the total amount of work to perform so that all processors complete their assigned tasks at the same time. One measure of load imbalance is the ratio of the difference between the finishing times of the first and last processors to complete their portion of the calculation to the time taken by the last processor.
locality (n.) The degree to which the computations done by a processor depend only on data values held in memory that is close to that processor, or the degree to which computations done on a point in some data structure depend only on values near that point. Locality can be measured by the ratio of local to nonlocal data accesses, or by the distribution of distances of, or times taken by, nonlocal accesses. See also halo, stencil.
locality of reference (n.) the observation that references to memory tend to cluster. Temporal locality refers to the observation that a particular datum or instruction, once referenced, is often referenced again in the near future. Spatial locality refers to the observation that once a particular location is referenced, a nearby location is often referenced in the near future.
lock (n.) Any device or algorithm whose use guarantees that only one process can perform some action or use some resource at a time.
logarithmic cost criterion (n.) cost criterion that assumes the cost of performing an instruction is proportional to the length of the operands. The integer N requires at least log N + 1 bits of memory, hence the name.
logic (n.) the branch of mathematics that investigates the relationships between premises and conclusions of arguments.
logic program (n.) a set of Horn clauses, or procedures.
logical inferences per second (n.) See LIPS.
logical time (n.) Elapsed time as seen from within processes. This may differ from clock time, because processes can block or be suspended during multitasking and because they can run at different speeds. The logical times of events only define a partial order on those events.
loop unrolling (n.) A compiler optimization technique in which the body of a loop is replicated L times, and the number of iterations of that loop reduced by a corresponding factor L. By lengthening the basic block inside the loop, this can increase the scope for vectorization and other optimizations.
loopback test (n.) A circuit test at any device which will tie the transmit data to the receive data in order to apply a signal and receive the data back for interpretation.
loose synchronization (adj.) A program running on a concurrent computer is said to be running in loose synchronization if the nodes are constrained to intermittently synchronize with each other via some communication. Frequently, some global computational parameter such as a time or iteration count provides a natural synchronization reference. This parameter divides the running program into compute and communication cycles. See also synchronous.
low-order interleaving (n.) a memory interleaving based on the low-order bits of the address.
lower triangular matrix (n.) a matrix with no nonzero elements above the main diagonal.
LU decomposition (n.) a technique where a matrix A is represented as the product of a lower triangular matrix, L, and an upper triangular matrix U. This decomposition can be made unique either by stipulating that the diagonal elements of L be unity, or that the diagonal elements of L and U be correspondingly identical.