Concurrency across component selects would be interesting, but that is
_really_ not there in the mysql codebase. :)
eric
On Feb 18, 2009, at 11:18 AM, Rick James wrote:
> In some cases you can get a big gain by using multple CPUs?
>
> * Seriously disk-bound: No advantage to merge.
> * No I/O needed: Speedup should be on the order of
> MIN(#cpus, #subqueries) - 1
> (The "-1" is because of the merge, itself).
> * Busy system (other threads competing for CPU) -- less savings
> * Mixture of CPU and I/O -- varing savings.
>
>> If the streams are not
>> ordered, you don't need interleaving at all and can execute
>> subqueries
>> sequentially, that's what existing code is doing. You only
>> get a gain if
>> you don't accumulate the results, and that's the only reason,
>> as far as
>> I can see, to implement this merge algorithm.
>>
>> Regards / Mit vielen Grüßen,
>> Sergei
