The problem is if you immediately jump up and declare some solution
that uses threads is just another case of "thread-mania", without even
properly reading and analysing the proposal.
> Thanks to god, I have had to work on some good O/Ses.
> About I/O completion, for example, I do like how TANDEM/GUARDIAN handles
> this.
>
> If I remember correctly, this works as follow:
>
> - User program associates a TAG to each request.
> - When user program wants to wait for completion it calls the
> AWAITIO() system service.
> - On return, the O/S indicates the TAG corresponding to the request that
> is complete.
> - Etc ...
>
> I think that the VMS semantic for I/O completion is a lot too complex
> to implement on UNIX systems, but the GUARDIAN semantic that is as
> efficient, IMO, could be implemented without conflicting too much with
> UNIX semantic.
I prefer the "event queue" scheme, where FDs that have new activity
send a message down a pipe which contains the FD number.
> I will look into your io-event proposal, and will tell you my opinion about.
Thanks.
Regards,
Richard....
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu