People are always looking for more effective cache mechanisms, particularly for ways of reducing the miss penalty; for example, by means of annex caches or victim caches. A student makes the following suggestion. Not all data is the same. Some numbers are used more frequently than others, particularly small numbers. So, why not arrange the cache so that if there are two candidates for eviction, the higher value is ejected first?
What will be an ideal response?
The question is based on faulty reasoning. It is true that the distribution of numbers is not flat and small
numbers are used far more frequently than large numbers (which is why most processors are happy to use 8?bit
or 12?bit constants). However, there is a difference between the use of numbers and their reuse. A number is
cached because it has already been used.
The argument fails because not all values that are cached are numeric. For example, if people’s names are
cached, it may be that higher values are more likely to be used than lower values.
In order to test this thesis it would be necessary to create a cache simulator and test whether ejecting the
larger candidate, when there were two possible candidates for ejection, has any effect on the average
performance of the cache.
You might also like to view...
Action
a) method, parameters. b) name, params. c) label, names. d) None of the above.
Google+ is tied into many of Google's other services, including Gmail and the Google search engine, for a unified user experience
Indicate whether the statement is true or false
The precedence of an operator establishes its priority relative to all other operators.
Answer the following statement true (T) or false (F)
A monitor's screen is divided up into millions of tiny dots, called:
A. bits B. bytes C. cursors D. pixels