RE: [SSI-devel] clusterwide utmp

[Walker, B. J <bru...@hp...>]

Aneesh,
    Sorry we didn't mention this earlier.  The plan is to use devpts,
partly because it is the preferred method in 2.5.   For naming, cfs will
be stacked over it.

-----Original Message-----
From: Kumar, Aneesh (Digital GlobalSoft)=20
Sent: Tuesday, July 22, 2003 10:25 PM
To: fr...@ex...
Cc: Walker, Bruce J; Byrne, John (OpenSSI clusters); ssic-linux-devel
Subject: Re: [SSI-devel] clusterwide utmp


On Wed, 2003-07-23 at 10:20, Frank Mayhar wrote:

>=20
> I think so, yes.  This is why you use a name server.  You may need=20
> something like a key service that lives on some particular node, but=20
> however it's implemented, you'll have one name server that allocates=20
> the names in /dev/pts.  So on node 1 you'll have /dev/pts/0, on node 2

> you'll have /dev/pts/1.  Walking /dev/pts would probably traverse all=20
> nodes in the cluster (much like walking /proc in Unixware).
>=20

Okay. If we are going with the above. I would say we should modify
devpts rather than hacking devfs . That would pay us later.=20

What do you say john ?

-aneesh=20