Check out the new USENIX Web site. next up previous
Next: Varying Inter-cache Bandwidth Up: Results Previous: Two Cache Hierarchy

Differing Cache Sizes

Figure 11: On the x-axis is the size of the $ C_1$ (higher) cache. $ \vert C_1\vert+\vert C_2\vert = 100K$ blocks.
\begin{figure*}\begin{center}
{\small Varying Relative Cache Sizes in a Two Cach...
...}
\epsfig{file=epsmixed/P1-art.eps, height=2.1in, width=3.25in}
}\end{figure*}
In Figure 11, we vary the relative size of the $ C_1$ and $ C_2$ caches from $ 1:9$ to $ 9:1$ while keeping the aggregate cache size equal to $ 100K$ blocks. We present average response time and inter-cache traffic (assuming unlimited bandwidth and free demotions) for the trace P1 (other traces have similar results). We observe that PROMOTE variants have consistently better response times than the DEMOTE variants across the entire spectrum of relative sizes. The average response time for plain LRU peaks (implying that the hit ratio is the lowest) when the two caches are of the same size. This confirms that the most duplication of data happens when the caches are of comparable sizes. For all the other algorithms, the average response time decreases as the size of the $ C_1$ cache increases as more hits occur at the highest cache.

As before, we observe that the DEMOTE variants invariably consume $ 2$x bandwidth when compared to the PROMOTE variants.


next up previous
Next: Varying Inter-cache Bandwidth Up: Results Previous: Two Cache Hierarchy
root 2008-01-08