pmbw is a set of assembler routines to measure the parallel memory (cache and RAM) bandwidth of modern multi-core machines. Memory bandwidth is one of the key performance factors of any computer system. Today, measuring the memory performance often gives a more realistic view of the overall speed of a machine than pure arithmetic or floating-point benchmarks. pmbw contains a set of very basic functions which are all hand-coded in assembler to avoid any compiler optimizations. These basic functions are modeled after the basic inner loops found in any data processing, sequential scanning and pure random access. Any application will have a memory access pattern which is somewhere between these two extremes. The current version of pmbw supports benchmarking 16-, 32-, 64-, 128-, or 256-bit memory transfers on x86_32-bit, x86_64-bit, and ARMv6 systems.
sdb is a really simple program to benchmark disks. It only writes/reads data sequentially to/from a file and calculates the performance of this action. A memory buffer is used for input/output to get values close to the real performance of the disk or RAID system. You can use flags like O_DIRECT and O_SYNC to avoid buffering by the operating system. It is similar to dd, but there are some differences, since dd was not created for benchmarking a disk. It is possible to initialize the buffer with random values before writing them to the disk or a file on the disk. It is a simple but effective program to get the maximum speed of a disk. Due to the simplicity, you will get the results much faster than with a more complex benchmark.
MPIBurn is an MPI unicast benchmark. Modern inter-connect hardware handles broadcast messages with a different algorithm than that used for unicast messages. Just benchmarking broadcasts or N to N+1 messaging may not detect inter-connect malfunctions, so MPIBurn benchmarks any possible unicast connection while assuring to involve as many connections in parallel as possible. After each round, the peers are scrambled to ensure that the next round has completely different connection groups.
Likwid is a set of easy to use command line tools for Linux. It supports programmers in developing high performance multi-threaded programs. "Likwid" stands for "Like I knew what I am doing". It contains the following tools: likwid-topology, which shows thread and cache topology; likwid-perfctr, which measures hardware performance counters on Intel and AMD processors; likwid-features, which shows and toggles hardware prefetch control bits on Intel Core 2 processors; likwid-pin, which pins a threaded application without touching its code (it supports pthreads, Intel OpenMP, and gcc OpenMP), likwid-powermeter which prints the Turbo mode steps and measures energy consumption on supported Intel processors, and likwid-bench, a low level benchmarking framework. It works with any standard Linux kernel. Likwid is lightweight and adds no overhead during measurements.