G
George Potemkin
Guest
> Also, the scatter really only works when you have a fair amount of data. With few records and few blocks it doesn’t change much. Now I understand. Indeed it was my mistake: I compared the scatter factors for a table with two data blocks. In first case each block stored 4 records, in second case - only one record. But in the best case a block could store 8 records. So in both cases the ratio of the real block count to the optimal one was 2:1 and that is why the scatter factor was 1.3 in both cases. I run new tests with 10.2B as well as with 11.6. The scatter factors reported by dbanalys were equal the values calculated by the following formula: ASSIGN vMaxRPB = (vDbBlockSize - 64 - vCreateLimit) / (vMeanRecSize + 2) vMaxRPB = MIN(vMaxRPB, vAreaRPB) vRealRPB = vRecCount / vBlockCount vScatterFactor = 1 + LOG(vMaxRPB / vRealRPB, 10) . /* ASSIGN */ I'm not sure that dbanalys uses exactly same constants to estimate vMaxRPB as in the formula above. But scatter factor is reported with low precision so the real values of these constants don't seem to matter. Conclusion #1: the scatter factor can be used to estimate the number of blocks that store the table's records resided in type two storage area. Note: an object block stores the number of clusters owned by object. But scatter factor is based on the block count rather than on cluster count. Conclusion #2: the large scatter factors for the tables in SAT2 seem to mean that the tables have long RM chains. Long RM chain is a "bad thing" and it leads to the performance degradation for the reading. Hence the scatter factors larger than 1.0 need your attention even in SAT2. Best regards, George
Continue reading...
Continue reading...