Re: Needful changes for cluster operation
Mark Stier <kalium@gmx.de> writes:
> Here's a list of things needed to be changed to make Debian really
> cluster-operatable - at least as far as I know. I'm assuming that it's a
> client-server cluster to which diskless workstations can be attached
> to. I'm (therefore) additionally assuming that there's one server
> carrying all system software, all data and all client
> configurations. Client's harddisk -- if available -- may be used as
> nfscache, for swap, for "/tmp" and for booting.
That is the boring case, as far as I'm concerned.
> 1. The main problem I stepped onto are the daemons and
> SysVInit. Solution: cluster-wide configuration data containing
> configuration data for all machines. The configuration data sets may be
> linked to the ethernet addesses. This method would be "clear" and no
> changes related to NFS daemons would be necessary (NFS servers sending
> different files to different machines depending on IP or ethernet
> address).
See "diskless" package - it does that already. And quite well, at least
year or so ago.
> 2. Setting up a cluster for different platforms requires two or more
> versions of each binary. Would a Debian system work properly if we split
> binary dirs from /usr/bin to /usr/i386/bin and /usr/m68k/bin etc. and set
> BIN/LIB variables for each machine accordingly?
Ugh. That sounds like.. brr. I can't even think about it. Why not have
separate server for each binary arch and one server with shared data? Of
course, if you intend to build depressing "one client of each arch,
starting with PalmLinux" model it might not work..
> What's your opinion?
>
> Regards,
>
> Mark
-Markus
> --
> Men occasionally stumble over the truth, but
> most of them pick themselves up and hurry off
> as if nothing had happened.
> -- Winston Churchill
--
"One World, One Web, One Program" - Microsoft Promotional Ad
"Ein Volk, Ein Reich, Ein Führer" - Adolf Hitler
We don't see a similarity. We really don't.
Reply to: