On Tue, 30 Nov 1999, Stew Forster wrote:
> What memory hit rate I can see is irrelevant. That is entirely algorithm
> dependent. What is important is that object re-access patterns more
> closely follow real-world scenarios, rather than some arbitrary forced
> memory-miss scenario.
You probably should not complain about irrelevant stuff then. :) If you go
back where this thread has started, you will see that the original question
was the parameters for Zipf distribution that we should use. One way of
estimating that parameter is using known memory hit ratio (see Pei's
e-mail), but there are other ways.
> This then allows for better memory caching algorithms to strut their stuff,
> while others may only be capable of much less. The point is, if the load
> follows more real-world patterns, then caches that are tuned for that
> behaviour can perform their intended function, rather than being reduced
> to a disk-only test.
This is all very true, but things like "more real-world patterns" do not
help us to move forward with the workload specs. Everybody agrees that the
patterns should be more real-world, that is not the issue.
The current framework is very simple: you are given a workload that can use
a [sliding] Zipf(alpha) popularity distribution. We are asking for vendors
help in determining the value of alpha. The value of zero yields a uniform
distribution. The value of one is obviously not realistic. Somewhere
between 0 and 1 is the sweet spot that you can help us to find based on
your observations and all the research you are doing.
If you think alpha is pretty much constant for a large number of workloads,
great. If you think alpha is request rate dependent, please give us a
table/function instead of a single value.
Finally, if you consider the framework itself to be invalid or unrealistic,
the time to suggest better models for the second bake-off has passed. We
would be happy to consider new models for the future tests, of course.
Thanks,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:09 MDT