Scaling: Future Improvements
Message ìheapî, continued
- Virtually eliminates all sync. meta-data updates
- Could potentially be combined with previous single-instance-store ideas
- Does increase maintenance overhead
Of course, now youíve just thrown out all that previous experience with well-tested filesystems code.
However, since you donít need everything that a full filesystem comprises, and you donít need everything that a full-blown RDBMS provides, you can pick and choose the parts that work from both camps, and you should still be able to put something together that will work reasonably well.
After all, there are plenty of INN and Diablo news servers out there that are handling a full 200GB+ full feed these days with similar technologies, and they seem to work pretty well.
So long as you build in enough redundancy that you can recover the database portion should it get corrupted, it seems to me that this would be a good balance between storing everything message-per-file in a filesystem, and storing everything as BLOBs in a database.