... We describe a strategy for optimal memory and work distribution. ... A bodies data distribution ranging from contiguous blocks (coarse grain: cdir$ shared pos(:block,:)) to a fine grain distribution (tf ) (cdir$ shared pos(:block(1),:)) was adopted. ... The medium grain bodies data distribution (bm) is obtained using a block factor N = Nbod /2 N $PE s; i.e., each PE has two data block of bodies properties residing in the local memory, each block having a close bodies set. ... 0 Plp 1). ...