Various visualization methods based on out-of-core technology have been introduced to deal with large datasets whose sizes exceed the capacity of local memory on a single desktop PC. However, the visualization of pathlines for huge, time-varying datasets on a single machine is not feasible with naive methods due to the low hit ratio of a general cache system based on LRU. In this paper, we propose a novel cache management strategy which relies on the processes of data pre-fetch and cache evacuation. We also evaluate the proposed method using a real-world flow dataset. Our experiments show that the proposed pre-fetch method increases the cache hit ratio whereas an evacuation strategy decreases the re-fetch ratio.