So far only one of the representations of preview state, CumulativeInclusionRatio, is used to illustrate the concept and representation of the preview state. Jumpshot-4 actually uses several different representations of preview state. All these representations are based on 2 ratios stored in SLOG-2 file. They are called Inclusion Ratio and Exclusion Ratio. Inclusion ratio is computed without taking into account of the nesting order of the states. States which are either nested inside or enclose other states contribute equally to the inclusion ratio. The end result is that the sum of all inclusion ratios from all state categories in a preview state could easily be larger than 1. On the other hand, exclusion ratio is specifically computed to exclude the overlap of the nested state from the enclosing state. Therefore the sum of exclusion ratios of all state categories in a preview state is guaranteed to be less than or equal to 1.
The motivation of computing these 2 ratios is to satisfy two opposite needs of preview state. If you are a MPI application developer and you have put a lot of user-defined states in your SLOG-2 file through either MPE or AIX's PCT utility, you most likely would be interested in the profiling information of the user-defined states which enclose MPI states and other user-defined states. In this case, inclusion ratio will be very useful. Because inclusion ratios of user-defined states usually dominate all state inclusion ratios, including those of MPI states. Therefore, the inclusion ratio highlights the outermost enclosing states even at high preview level. On the other hand, if you are a MPI implementor or are interested in the low level MPI networking overhead, you are most likely interested in the profiling information of MPI and its internal calls. Exclusion ratio will come in handy. Exclusion ratios for the innermost nested states, i.e. MPI states, tend to dominate all state exclusion ratios. So the exclusion ratio highlights the innermost nested states at very high preview level.
![]()
|
Figure 2.6 shows a typical zoomed in view of some nested states. In this view, the yellow states are deeply nested in the navy blue state which is in turns nested in the orange state. The pair of green lines mark the region where a preview state is being created for.
|
The inclusion and exclusion ratios are computed for the region marked by the pair of green lines and are shown in Table 2.2. The table shows that the most dominant state among all inclusion ratios is the orange outermost state, but the most dominate state among all exclusion ratios is the yellow innermost state which is the least dominant state in inclusion ratios. One obvious observation is that the inclusion and exclusion ratios of the innermost state category are the same.
![]()
|
With data computed in Table 2.2, various different preview displays can be drawn and are shown in Figure 2.7. All colored strips inside the preview state will be drawn proportional to the height of the preview state. For instance, if the ratio of the category for the strip is 0.9, the corresponding colored strip will occupy 90% of the preview state's height. The statement is true for all preview state display except CumulativeInclusionRatio which could have its total sum of ratios in exceed of 1.0 especially when the slog2 file is highly nested. First consider the CumulativeInclusionRatio and CumulativeExclusionRatio views, i.e the first and the third ones from the top in the figure. Notice that yellow state is least important in the top CumulativeInclusionRatio view, but becomes most significant in the third CumulativeExclusionRatio view. Since the sum of all inclusion ratios is larger than 1, in this case, the sum is 2.3, the CumulativeInclusionRatio view reweights all ratios to fill up the preview box. Strictly speaking CumulativeInclusionRatio view cannot be used to compare different preview states because of the arbitrary rescaling2.6. If one is interested in the comparison of inclusion ratios across different preview states, OverlapInclusionRatio view can be used instead. OverlapInclusionRatio view draws all inclusion ratios proportional to the height of the preview state but in an overlapping way, i.e. draw them in decreasing inclusion ratios order and stack one on top of the other, sort of like nested state. The overlap view of exclusion ratios is OverlapExclusionRatio view which is shown at the bottom of Figure 2.7. OverlapExclusionRatio view draws exclusion ratios exactly the same way as OverlapInclusionRatio. In general, overlap view cannot fill up the full height of the preview state. This is apparent in OverlapExclusionRatio view in 2.7 where the white bordered box indicates the full height of the preview state. The white bordered box is necessary in comparing the ratios across different preview states with respect to the preview states' duration. However, the white bordered box can sometimes be confusing, because whatever in the back of the preview state can show through the empty space within the white bordered box. In that case, the bordered box can be turned off by selecting Empty in PREVIEW_STATE_BORDER in Preference window.
For the sake of comparison and continuity with our preview discussion, the CumulativeExclusionRatio view of Figures 2.1 and 2.2 are shown in Figures 2.8 and 2.9 respectively. The CumulativeExclusionRatio view does provide an extra dimension of information when compared to its inclusion ratio counterpart at the expense of being a bit more complicated to digest visually.