From Parawiki

Speed-up results

Speed-ups have been measured on two systems:

Cluster consisting of 3 quad-processor machines (IBM RS6000 pSeries 630). Specification of each machine:
- CPUs: 4 x IBM PowerPC POWER4, 1000 MHz
- Operating system: AIX 5.2.0
- Main memory: 16 GB (2 machines), 2 GB (1 machine)
Cluster of 38 dual-processor machines. Specification of each machine:
- CPUs: AMD Opteron 248, 2.2 Ghz
- Operating system: Redhat Enterprise Linux WS 64 Bit
- Main memory: 4 GB

In the following tables SpdUp is the speed-up in comparison to the same parallel implementation with only one worker (in worker-coworker programs) or one slave (in master-slave programs).

SpdUp(S) is the speed-up in comparison with the serial implementation.

Back to Parallel Virtual Machine (PVM)

Speedups with PVM (one 4 CPU machine)

Please note that the machine just has 4 CPUs. So normally speed-ups for elastic and thresh should only be achieved with 3 (= 4 processes) or less. In constrast to this assumption we get an almost ideal speed-up even with 5 processes (4 slaves). The reason for this behaviour possibly is a better usage of the CPU and a better load balancing if more than process is running on one CPU. Another reason is that the machine on which we ran our tests is used by many people at the same time. Each process gets only little time slices for execution. When we run more processes we might get more time slices from the operating system scheduler and our program runs faster.

Speedups with PVM (one 2 CPU machine)

Speedups with PVM (three 4 CPU machines)

Annotaion: With most problems no better speed-ups can be achieved. The reason for those bad results might be the increasing communication costs when sending data over the network instead of sending data without usage of the network in one machine.

Only those programs with a lot of calculation have are faster than serial execution.