Re: mempool patches

On Tue, 25 Aug 1998, Henrik Nordstrom wrote:

> Which also tells us that there is no point in supporting freeing of
> pooled memory, as virtually nothing is freed either.

Freeing memory is important in these two cases:
        - reconfigure with a lower mempool limit;
        - supporting mempool limit (other than infinity).

If we decide that infinite mempool limit is what everybody should use, then
the code should be 90% rewritten. The current code is simple. The "infinite"
code will be simple as well. IMO, the patch comes in-between trading the code
clarity and versatility for questionable improvement in memory fragmentation.

IMHO, a performance patch that makes code more complex should be rejected
unless it gives more than 10-20% improvement. This is, of course, an
idealistic approach.

Alex.

P.S. There were plans to "pool" virtually all dynamic memory allocations.
This should be kept in mind when deciding if we need a mempool limit or not.
Received on Tue Jul 29 2003 - 13:15:53 MDT