This paper compares the trace-sampling techniques of set sampling and
time sampling. Using the multi-billion-reference traces of Borg et al.
, we apply both techniques to multi-megabyte caches, where sampling is
most valuable. We evaluate whether either technique meets a 10% sampl
ing goal: a method meets this goal if, at least 90% of the time, it es
timates the trace's true misses per instruction with less-than-or-equa
l-to 10% relative error using less-than-or-equal-to 10% of the trace.
Results for these traces and caches show that set sampling meets the 1
0% sampling goal, while time sampling does not. We also find that cold
-start bias in time samples is most effectively reduced by the techniq
ue of Wood et al. Nevertheless, overcoming cold-start bias requires te
ns of millions of consecutive references.