| OCR Text |
Show CHAPTER 5 RESULTS & ANALY SIS 5.1 Introduction This chapter descr ibes various tests that were performed on diffe rent POSs and the resu lts that were obtained . T' ll(-' a nalysis of t hese results draws upon the available lite ratu re, the code for the POSs, resoiirce usage by various processes, and specific tests. One of tbe major co ncerns in all these tests was u11 iformity in implementation . To e nsure uniformity, all three implementations were clone by one p erson , the aut hor, and an effort was made to keep the software des ign cons istent. Within t lw cons! ra in t of til t• OVlTctil design . t.IIl' <Uithor t ried to writ(-' each implt'fn(jnt ctLi(J!I t.o oiJLain t he lwst p(-'rfo1·mance for Llw p<uticular POS. T he (esLs that were performed differ from most of t he benchma rks for OODBMSs s11ch as, 001 , 007, Hyperrnoclel, and Vliss Be ncbmark . This is because a ll these lwnchmarks assume an " in teract ive" model for CAD applications. Alpha_l , on the otl1e r hand, interacts with P OSs in a "batch" mode, i.e., all the data is loaded in at t he start of t he session and dllmpecl at the e nd of the session . All time IlW<-tSII JTJ1H' JI Ls njportcd here are '·'cold" res 1tlts. To prevent cach ing from (~arli e r t('s(s Crom having an eHecL on tim ing, t he database was closed before each new tesL and the file bufFer cache was flu shed by a process which read in a la rge file Ill 0 nou-seqi tcuLial manner. For a ll th ree POSs, the database was ini t ia Lized in a st·parate process. so t he illiLialization wa.s not a part of tbe measurements. |
