Hi Stefan, Stefan Strasser wrote:
writing 512 bytes, aligned to 512 bytes, without O_CREAT, when the file already exists, brings the desired results 2 seconds with much less disk usage. that's some set of conditions.
I would be interested to know how this compares to mmapping the file and then writing to memory and calling msync(). I while ago I did some work on a transactional persistent object library for "write mostly" data. Since it didn't need to support efficiently erasing or modifying existing data it could use a pure log structure with transactions implemented just by atomically adjusting the end-of-data offset. I did this all via mmapped files and the performance seemed good enough, but I never benchmarked it against e.g. MySQL. Actually it has been on hold since I discovered how badly it performs over NFS... Phil.