Table III further classifies the techniques based on the fraction of cache that is locked
(i.e., full or partial). In the full cache locking approach, all accesses to unlocked blocks
lead to cache misses, which harms performance. By comparison, partial cache locking
can provide better performance by virtue of carefully deciding the locked fraction of
cache based on cost-benefit analysis. However, from the perspective of predictability, in
partially locked caches, CRPD analysis and static cache analysis are still required for
the unlocked portions. In contrast, full cache locking does not require cache modeling
for WCET analysis, and hence, these techniques can achieve tight WCET estimates.
These techniques assume complete cache-bypass as the baseline and improve upon it by
locking the memory blocks along the worst-case execution path, which is guaranteed to provide maximum WCET reduction over the baseline. It is clear, that these approaches
offer performance/predictability tradeoffs