This thesis is concerned with the problem of determining optimal storage allocation strategies that are sensitive to the nature of the processes that request storage. It is intuitively obvious that such a strategy would be better than one which has been decided upon a priori, and does not take into account the local behavior of the entities it deals with. Efforts have been made to characterize the storage usages of computer users. Based on the techniques of Time Series Analysis, stochastic modles are constructed for the VSLM, SLRUM and IRM models of Denning and Spirn, It was found that the storage requirements of processes essentially stay on a certain level most of the time. An entity, called holding probability, is defined for a process to denote the probability that this process will change its storage need. Most programs under our studies have holding probabilities between .5 to .2. Also, experience from the experiments has shown us that the mean working set size of process can be determined quite accurately. In the second part of the thesis, a storage allocation policy called DFSA is discussed. The assumption that WS size as a function of time is normally distributed has been corroborated by our study of program behavior, GPSS/36 simulation has been used for the purpose of providing information about the performance of DFSA as compared with the currently used policies, The results show that roughly speaking, we need only half as much as channel activity while achieving a throughput rate not worse than the currently used strategies. The user's response time, under DFSA, however, is larger by 3%.