| 
|
Topic: computer performance
topics >
computer science >
computer architecture
Topic:
efficiency
Topic:
interprocess communication
Topic:
massively parallel processors
Topic:
memory cache
Topic:
models of parallel computation
Topic:
multiple processors
Topic:
real time systems
Topic:
RISC computer architecture
Topic:
virtualized hardware
Subtopic: performance
Quote: portable API for performance tools using the on-chip performance monitoring hardware; measurements per-thread; PAPI [»muccP6_2005]
Subtopic: performance problems
Quote: off-the-shelf computers are as good a researchers' computers; can't buy high-performance computers to study what will be available in five to ten years [»myerB3_2000]
Subtopic: bandwidth, latency
Quote: periodically stop development for a week to analyze and resolve performance problems; otherwise minor performance losses will accumulate [»colwRP_2006]
Subtopic: capacity
Quote: bandwidth improves faster than latency; when bandwidth doubles, latency improves 1.2-1.4x [»pattDA10_2004] Quote: store data in large block sizes; balances bandwidth with latency [»pattDA10_2004] Quote: latency improvements also improve bandwidth, but not vice versa [»pattDA10_2004] Quote: use caching, replication, and prediction to improve latency [»pattDA10_2004] Quote: high latency, communication overhead, and limited bandwidth will remain problems for parallel processing [»cullDE11_1996] Quote: the LogP model of parallel computation uses latency, overhead, gap between successive sends/receives, and number of processors; fixed size messages and finite capacity network
Subtopic: memory
Quote: memory and storage capacity improves faster than bandwidth Quote: storage capacity doubles every year while access time improves at 10% per year; increasingly difficult to extract data [»grayJ6_2003]
Subtopic: size
Quote: the limiting factor for nonvectorized code is reading operands from memory; makes workstations comparable to supercomputers [»thomCD11_1993] Quote: memory latency is almost 1000 instructions; may be faster to recalculate than to fetch [»pattDA10_2004]
Quote: with 0.1 micron wires, unbuffered gates can only reach 0.36 mm at 1.2 GHz; constrains design into small regions
[»matzD9_1997]
Topic: computer architecture (46 items)
Topic: efficiency (96 items)
Topic: interprocess communication (29 items)
Topic: massively parallel processors (29 items)
Topic: memory cache (29 items)
Topic: models of parallel computation (33 items)
Topic: multiple processors (10 items)
Topic: real time systems (14 items)
Topic: RISC computer architecture (11 items)
Topic: virtualized hardware
(5 items)