P4
RFE idea from Dean Long in regards to the new CDS implementation after the upcoming permgen elimination work is integrated.
>>
>> For dump, it looks like the code is still allocating anonymous memory first and then writing to a file later. A slight optimization would be to create the archive file early, then map the zeroed out file instead of anonymous memory.
>
> I didn't know how to do this. You mean create a shared map file and it would just write the data in that?
Yes, dump would create a file and map it the same was restore opens a file and maps it. Then there would be no more writing into the file with write(2) but instead direct memory writes.
>>
>> For dump, it looks like the code is still allocating anonymous memory first and then writing to a file later. A slight optimization would be to create the archive file early, then map the zeroed out file instead of anonymous memory.
>
> I didn't know how to do this. You mean create a shared map file and it would just write the data in that?
Yes, dump would create a file and map it the same was restore opens a file and maps it. Then there would be no more writing into the file with write(2) but instead direct memory writes.