[Beowulf] Woodcrest and bad result

[Mat Agar]

I have 5 node cluster with two 5120 Xoen dual core (each node has 4 
cpus) and 2GB ram ,on a single host when I run Intel linpack_9.1 
(xlinpack_xeon32) with OMP_NUM_THREADS=4, it reports about 17 GFLOPS but 
when I run HPL with "mpiexec -n 4 -host node-1 ./xlinpack" it achieves 
about 4 GFLOPS . Also In network situation, I can not achieve to desire 
speed not only by Intel pack nor by HPL. For example with Intel for 4 
nodes the report is about 23 GFLOPS
 and HPL result is awful. Dose any body know what is wrong?