Khamis Siksek wrote:
10 rows is just an example, but replacing 1 MB for 100 bytes change
every time there is an update the overhead is more than 99%.
How about using the same methodology that xdelta is using to do the
binary patching, also the client doesn't have to know anything about the
offset, he just need to call a function "Update" with the right key, the
server should handle the patching not the client.
If you're sending the whole dataset to the server, it's no more overhead
to copy over the data. It'd be a lot more computationally intensive to
do an xdelta on what to copy around in RAM. Also, 1MB is the upper limit
for slabs, so they shouldn't get much bigger than that :)
-Dormando
