[RFC] Proof-of-concept for netlink listener for mdev -i

[Harald Becker]

On 20.03.2015 21:28, V.Krishn wrote:
> The concept of a simple netlink reader daemon seems nice.
> But would seem extra to having another reader that reads this reader.
> The concept of whole /sys tree is being re-done, even if its just not
> visible(memory). I could add/suggest few changes.
> Please consider concept/flow below as theoretical.

1) Your question is about the principal of putting all together in one 
logical process, or splitting operation into different threads. Most 
likely will modern multi core processors benefit from splitting into 
threads, as those may run on different cpu cores, but also single core 
machines shall gain benefit, as the pipe has buffering purposes and 
allow for more parallel operation.

2) The pipe is pure memory operation, which will never access any 
physical disk, so we do memory writes, even faster than writing anything 
into /sys/...

3) The reader is started on demand, that is this process is fired up 
only when events arrive and it shall exit, when idle. So the bigger part 
drops it's resources giving more to the system for other operation. Only 
the small netlink reader part stays in memory.

4) Putting small, functional simple operations together in a pipe, is 
the Unix way. netlink is a long lived daemon, but concentrating on it's 
single job, to read from netlink, reformat the messages into textual 
format, then write them to stdout, make it more compatible to the Unix 
philosophy (IMO). And it make this piece of software even compatible 
with other textual Unix tools, which can give extra benefit, when you 
like to do need things, like display all incoming event messages on a 
console.

--
Harald