Change the smallest quantum we're willing to allocate to any single
bufferevent in a group for reading or writing at a time.
The rationale is that, because of TCP/IP protocol overheads and kernel
behavior, if a rate-limiting group is so tight on bandwidth that you're
only willing to send 1 byte per tick per bufferevent, you might instead
want to batch up the reads and writes so that you send N bytes per
1/N of the bufferevents (chosen at random) each tick, so you still wind
up send 1 byte per tick per bufferevent on average, but you don't send
so many tiny packets.
Change the smallest quantum we're willing to allocate to any single bufferevent in a group for reading or writing at a time.
The rationale is that, because of TCP/IP protocol overheads and kernel behavior, if a rate-limiting group is so tight on bandwidth that you're only willing to send 1 byte per tick per bufferevent, you might instead want to batch up the reads and writes so that you send N bytes per 1/N of the bufferevents (chosen at random) each tick, so you still wind up send 1 byte per tick per bufferevent on average, but you don't send so many tiny packets.
The default min-share is currently 64 bytes.
Returns 0 on success, -1 on faulre.