"Bruce Momjian" <bruce@momjian.us> writes:
> Gregory Stark wrote:
>>
>> Is this still testing with all data fitting in RAM?
>
> Yes. Having things out of RAM is going to make access even slower, but
> it is going to allow the heap to be in RAM more often.
It would let us measure the actual impact of TOAST. The largest negative
effect of which is to destroy the sequential access pattern and the positive
effect is as you say to increase the cache effectiveness on non-toasted data.
The cpu costs are insignificant so testing in-memory cases misses the point.
-- Gregory Stark EnterpriseDB http://www.enterprisedb.com