[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: re[2]: Clusterwide pids



> 
> Let us summarize a bit:
> -most people agrees we need unique CPID even if the process is something we 
> know is not going to migrate as init. 
> -that CPID will be node number + local pid

process creation node number + process creation uniquifier

> - We don't want to recompile userspace apps so we do our own system calls, 
> which  maybe :
> 	* getcpid to get the ¿64bits? with the standard getpid returning the low 32 
> bits of the CPID
> 	* getnode(pid) with 0 if process if local , nonzero otherwise.

I certainly don't agree with this.  pid's are 32 bits so having both node number
and uniquifier is quite feasible (have been doing it for a dozen years).
This way it is all transparent, which, in the case of SSI clustering, is key.

> Then we have a question left: ¿How we manage to get unique CPIDs ?
> What number node a machine must have, ¿read it from /etc? ¿read it a la HDLC?

The issue, as I mentioned earlier, is when/how does a node get a node number.
The DLCP idea is interesting.  Another option is something in lilo or ram disk.  
Specifying it must be in /etc is bad for those of us with a single root.
I like the idea of a "cluster" system call.  One of the subcommands of it would
be set your node number.  Different cluster implementations could gather that
number from different places.

bruce

> 
> i have missed something?
> 
> 
> --
> 
> Jordi
>    Student of Spain 
> 	  
> 
> Linux-cluster: generic cluster infrastructure for Linux
> Archive:       http://mail.nl.linux.org/linux-cluster/
> 

Linux-cluster: generic cluster infrastructure for Linux
Archive:       http://mail.nl.linux.org/linux-cluster/