Notice that TPC-D's prepurge hazard tends to be much lower than that of the production workloads.
TPC-D stands out clearly among the workloads in that it exhibits very little intertransactional reuse of sequentially accessed pages.
Observe that TPC-D is unlike the production workloads in that prepurging, even with prefetching, increases the miss ratio.
TPC-C and TPC-D lie within this very broad spectrum, with 6.9:1 and 45.0:1, respectively.
Observe also that the write miss ratio for TPC-C, and especially TPC-D, is much higher than that of the production workloads.
Among all the workloads, TPC-D has the highest write miss ratio and is the least responsive to more relaxed age limits.
For both the production workloads and TPC-D, only about 20% of the pages in the buffer pool are dirty.
The same can be said for TPC-D. On the other hand, the difference in write miss ratio for TPC-C with relaxed age limits can be more than 50%.
We also analyze the reference characteristics of the de facto standard benchmarks for both online transaction processing and decision support systems, namely TPC-C and TPC-D. Comparing the reference behavior of these benchmarks with that of the production workloads, we find that, for the most part, the reference behavior of TPC-C and TPC-D fall within the spectrum of behavior exhibited by the production workloads.
TPC-C is generally consistent in this regard, but not TPC-D--indicating that strategies that only consider per-transaction reference behavior will perform disproportionately well for TPC-D.