FLYNN:COMPUTER ORGANIZATIONS 959 tem.This overabundance of execution bandwidth has 0R*/外 o呢水03s洲70源8o0LE become a major characteristic of large modern systems :terd 。 ●海 In highly confluent SISD organizations specificexecu- ro execute OH tion units are made independently and designed for 乃：pipelining factor% number of different maximum performance.The autonomy is required for opemttons which Beeution Bandvidth ease of implementation since from topological consid- caft be in process al oHe time r one utt erations an individual independent unit executing a particular class of operands gives much better perfor- Fig.11.Execution bandwidth mance on the class than an integrated universal type of unit 1.The other notable characteristic of the SISD trol unit point of view,the maximum decision efficiency organization is that each of these independent subunits is generated when the control unit has relatively simple must in itself be capable of large or high bandwidths proportions (i.e.,handles a minimum number of inter- since the class of operations which successive instruc-locks or exceptional conditions-as when J is small per tions perform are not statistically independent,and in instruction stream)and when it is being continually fact they are usually closely correlated;thus,for ex- utilized (a minimum idle time due to interstream or in- ample,systems like the CDC 6600 and the IBM 360/91 trastream turbulence or lockout).While shared resource both have execution facilities almost an order of magni- MI organizations have an advantage over the confluent tude in excess of the average overall instruction retire- SISD and the SIMD arrangements insofar as com- ment rate.Fig.11 illustrates the bandwidth concept. plexity of control,the control must be repeated M times. Given N specialized execution units,with execution An overlapped SIMD processor would probably have pipelining factor P:for the ith unit(the pipelining factor the simplest control structure of the three. is the number of different operations being performed by one unit at one time-it is the execution analog of con- Primary and Secondary Storage fuence),then if t is the time required to fully execute The optimization of storage as a resource is a rela- the ith type of operation,the execution bandwidth is tively simple matter to express.The task,both program and data,must move through the storage as expedi- execution bandwidth tiously as possible;thus the less time a particular task spends in primary storage,the more efficiently the Note the bandwidth is in reality a vector partitioned by storage has been used with respect to this particular the resource class i.We sum the components as a scalar resource., to assess gross capability.Notice that the shared re- In essence we have storage efficiency measured by the source MI organizations are,by definition,optimized for space-time product of problem usage at each level of the efficient use of the execution resources. the storage hierarchy.The "cost"of storage for a par- ticular program can be defined as Instruction Control storage cost=∑c5at The control area is responsible for communications in addition to operational control.The communication where i is the level of storage hierarchy,ci is the cost function is essentially a process of identification of per word at that level,s:is the average number of words operand sink and source positions.The control area of a the program used,and t;is time spent at that level. system is proportionally much larger in number of deci- While the preceding overly simplifies the situation by sion elements when confluence is introduced in the in- ignoring the dynamic nature of storage requirements, struction stream since additional entropy must be re- some observations can be made.The MI organizational solved due to possible interactions between successive structure,by nature,will require both task program instructions.These precedence decisions must be made and data sets for each of the instruction units to be to assure normal sequential operation of the system. simultaneously available at low levels of the hierarchy. The analog situation is present in many parallel systems. The SIMD arrangement requires only simultaneous For example,in SIMD those data streams which are access to the M data sets,while the SISD has the least activated by a particular instruction stream must be intrinsic storage demands.Thus in general identified as well as those which do not participate. Notice that elaborations for controls,whether be it due S1 MIMD 2 S1 SIMD 2 51 SISD to confluence or parallelism,basically resolve no entropy Thus the MIMD and SIMD must be,respectively,more with respect to the original absolutely sequential in-efficient in program execution t:to have optimized the struction stream (and hence none with respect to the use of the storage resource problem). The necessary hardware to establish these sophistica- ACKNOWLEDGMENT tions is strictly in the nature of an overhead for the The author is particularly indebted to C.Neuhauser, premium performance.From an instruction unit or con- R.Regis,and G.Tjaden,students at The Johns HopkinsFLYNN: COMPUTER ORGANIZATIONS 959 tem. This overabundance of execution bandwidth has oPBL/-E- O/- oi:s'#/&;OO:OO OPi Op, become a major characteristic of large modern systems. t;:tirnery,prijd3 D LE In highly confluent SISD organizations specific.execu- to execti 1 E tion units are made independently and designed for PI:mpipelfdzfactoret *. ~~~~~~~numherof zf7Ef maximum performance. The autonomy is required for opemtzimts whack &eeu1x-n,8dL ease of implementation since from topological consid- cavbeziprocessal o;ze tkmc rn one &'it erations an individual independent unit executing a particular class of operands gives much better perfor- Fig. 11. Execution bandwidth. mance on the class than an integrated universal type of unit [1]. The other notable characteristic of the SISD trol unit point of view, the maximum decision efficiency organization is that each of these independent subunits is generated when the control unit has relatively simple must in itself be capable of large or high bandwidths proportions (i.e., handles a minimum number of inter￾since the class of operations which successive instruc- locks or exceptional conditions-as when J is small per tions perform are not statistically independent, and in instruction stream) and when it is being continually fact they are usually closely correlated; thus, for ex- utilized (a minimum idle time due to interstream or in￾ample, systems like the CDC 6600 and the IBM\I 360/91 trastream turbulence or lockout). While shared resource both have execution facilities almost an order of magni- MI organizations have an advantage over the confluent tude in excess of the average overall instruction retire- SISD and the SIMVID arrangements insofar as com￾ment rate. Fig. 11 illustrates the bandwidth concept. plexity of control, the control must be repeated M times. Given N specialized execution units, with execution An overlapped SIMD processor would probably have pipelining factor Pi for the ith unit (the pipelining factor the simplest control structure of the three. is the number of different operations being performed by one unit at one time-it is the execution analog of con- Primary and Secondary Storage fluence), then if ti is the time required to fully execute The optimization of storage as a resource is a rela￾the ith type of operation, the execution bandwidth is tively simple matter to express. The task, both program N p and data, must move through the storage as expedi￾execution bandwidth = E tiously as possible; thus the less time a particular task j1l ti spends in primary storage, the more efficiently the Note the bandwidth is in reality a vector partitioned by storage has been used with respect to this particular the resource class i. We sum the components as a scalar resource., .. . T ~~~~~In essence we have storage efficien}c measured by the to assess gross capability. Notice that the shared re- g y y soue Igr space-time product of problem usage at each level of theefcienorganizatofsthe,by use ecuionisou the storage hierarchy. The "cost" of storage for a par- the efficient use of the execution resources.tiuaprgmcnbedfeds ticular program can be defined as Instruction Control storage cost cis t The control area is responsible for communications in addition to operational control. The communication where i is the level of storage hierarchy, ci is the cost function is essentially a process of identification of per word at that level, si is the average number of words operand sink and source positions. The control area of a the program used, and t, is time spent at that level. system is proportionally much larger in number of deci- While the preceding overly simplifies the situation by sion elements when confluence is introduced in the in- ignoring the dynamic nature of storage requirements, struction stream since additional entropy must be re- some observations can be made. The MI organizational solved due to possible interactions between successive structure, by nature, will require both task program instructions. Th-ese precedence decisions must be made and data sets for each of the instruction units to be to assure normal sequential operation of the system. simultaneously available at low levels of the hierarchy. The analog situation is present in many parallel systems. The SIi\ID arrangement requires only simultaneous For example, in SIAID those data streams which are access to the M data sets, while the SISD has the least activated by a particular instruction stream must be intrinsic storage demands. Thus in general identified as well as those which do not participate.II Notice that elaborations for controls, wrhether be it due SLIM to confluence or parallelism, basically resolve no entropy Thus the M\IMIID and SIWI\LD must be, respectively, more with respect to the original absolutely sequential in- efficient in program execution t, to have optimized the struction stream (and hence none withl respect to the use of the storage resource. problem). The necessary hlardware to establishl these sophistica- ACKNOWLEDGMENT tions is strictly in the nature of an overhead for the The author is particularly indebted to C. NTeuhauser, premium performance. Froml an instruction unit or con- R. Regis, and G;. Tjaden, students at Thle Johlns Hopkins