The two plots in the right column of Figure 6 display the rate (per second) of tracks being brought (or staged) to the cache in response to read misses or sequential prefetches. In the cache-sensitive configuration (top right), we observe that LRU Top is better than LRU Bottom for the higher load levels, while the opposite is true for the lighter load levels. In contrast, SARC is consistently better than both the LRU variants for all load levels.
To understand the importance of this metric, note that from a client's perspective, in the cache-sensitive configuration, as seen in the top, left panel of Figure 6, LRU Bottom consistently outperforms LRU Top by delivering a lower average response time. However, from a storage controller's perspective, LRU Bottom outperforms LRU Top for lower loads while the converse is true for higher loads. In contrast, SARC consistently outperforms both the LRU variants from both the perspectives. In other words, not only does SARC deliver a better performance to a client, it does so without unduly stressing the server.
Similar observations also hold for the cache-insensitive configuration, albeit, to a smaller extent.