Method for assigning job in parallel processing method and parallel processing method

In a parallel processing method for parallelly processing a plurality of jobs comprising communication processing between a host computer and a plurality of processors and calculation processing in said processors by said plurality of processors in a computer system comprising said host computer and said plurality of processors connected to said host computer through a common bus, a job assignment method in a parallel processing method characterized in that said plurality of jobs are continuously numbered by integers ranging from N (N is an integer) to M (integer greater than N), a job that is numbered as N+(M−J−1) is assigned to the processor as the next job if the number J of the job that has been assigned just before is not greater than (N+M)/2 and a job that is numbered as N+(M−J) is assigned to the processor as the next job if the number J of the job that has been assigned just before is greater than (N+M)/2, otherwise a job that is numbered as N+(M−J) is assigned to the processor as the next job if the number J of the job that has been assigned just before is less than (N+M)/2 and a job that is numbered as N+(M−J−1) is assigned to the processor as the next job if the number J of the job that has been assigned just before is not less than (N+M)/2 and respective jobs are assigned to processors in such a manner that an average value of ratios between said communication times and said calculation times falls within a predetermined value.

In a parallel processing method for parallelly processing a plurality of jobs comprising communication processing between a host computer and a plurality of processors and calculation processing in said processors by said plurality of processors in a computer system comprising a host computer and said plurality of processors connected to said host computer through a common bus, a job assignment method in a parallel processing method characterized in that said plurality of processors is divided into G (G is an integer not less than 2) processor groups in which the number of said processors is equal within the each group, said plurality of jobs is divided into G job groups in which the number of said jobs is equal within the each group, said processors groups and said job groups are associated with each other in a one-to-one relation and jobs within one job group are assigned to processors within corresponding one processor group, jobs in which sizes comprising said communication time and said calculation time and ratios between said communication times and said calculation times are approximate to each other belong to different job groups of said job group and jobs in which sizes comprising said communication time and said calculation time and ratios between said communication times and said calculation times are approximate to each other are assigned to processors in such a manner that assignment order of said jobs within said job groups differ from each other within said plurality of job groups.

In a parallel processing method in which molecular orbital calculation for calculating energy of molecule expressed by n (n is a positive integer) contracted shells is executed by a computer system comprising a host computer and a plurality of processors, with respect to all matrix elements F (I, J) of Fock matrix F expressed in the form of a sum concerning all of primitive basis functions contained in contracted basis functions I, J composed of primitive basis functions i, j in a calculation of a sum f(I, J)=f1(I, J)+f2 (I, J),

• an outermost loop is selected as a loop concerning a combination of said contracted shell R and T which can satisfy relationships of R≦n max (n max represents the maximum value of numbers assigned to n contracted shells) and T≦R,
• a second loop in the inside of said outermost loop is selected as a loop concerning said contracted shell S,
• a third loop in the inside of said second loop is selected as a loop concerning said contracted shell U,
• or said second loop is selected as a loop concerning said contracted shell U and said third loop is selected as a loop concerning said contracted shell S,
• a range of values which said contracted shell S can take is selected to be not greater than R, a range of values which said contracted shell U can take is selected to be not greater than R,
• a predetermined two-electron integral and a part of a predetermined Fock matrix element using the result of said predetermined two-electron integral are calculated at the inside of said third loop, said second and third loops are collected as one job unit and processing is assigned to said processor at said job unit, a parallel processing method characterized in that said jobs are formed in such a manner that a low number is assigned to a job whose orbital quantum number of said contracted shell R is large and a high number is assigned to a job whose orbital quantum number of said contracted shell R is small, wherein
• f1 (I, J) represents the sum concerning all contracted basis functions of products A1·P (K, L)·g (i, j, k, l) with a coefficient A1;
• f2 (i, j) represents the sum concerning all contracted basis functions of a product A2·P (K, L)·g (i, k, j, l) with a coefficient A2,
• g (i, j, k, l) represents the function value of a two-electron integral function g expressed by using primitive basis functions i, j, k, l which are components of primitive shells r, s, t, u contained in contracted shell R, D, T, U as indexes;
• and
• P (K, L) represents the element of density matrix P expressed by using a contracted basis function K composed of said primitive basis function k and a contracted basis function L composed of said primitive basis function I as indexes.