[rescue] A Question about Snap Servers

[velociraptor]

On Sat, 31 Jul 2004 17:15:36 -0400, Phil Stracchino
<alaric at caerllewys.net> wrote:
> On Sat, Jul 31, 2004 at 01:56:41PM -0500, Jonathan C. Patschke wrote:
> > On Sat, 31 Jul 2004, bitrot wrote:
> >
> > >does anyone have a nice SIMPLE way of setting up raid 1
> > >on solaris 9.
> >
> > 1) Create a tiny (10MB or 20MB) slice (I'll call this an mslice) on
> >    each disk for the metadb and a big slice on each disk to hold your
> >    data (I'll call this a dslice).
> > 2) Do metadb -a -f mslice1 mslice2 mslice3   # You really need three tiny
> >                                              # slices.
> 
> More is better.  Three is the *minimum*.
> 
> I quote:
> 
>             ... The majority consensus algorithm accounts for the
>      following: the system will stay running with exactly half or
>      more replicas; the system will panic when less than half the
>      replicas are available; the system will not  reboot  without
>      one more than half the total replicas.
> 
> In other words, three is the minimum number of metadb replicas with
> which it remains possible to boot the system after losing a replica.

Yes.  But performance degrades after some not so large number
of copies.  When I was working at $big_network_co, we usually 
limited the number of copies to about 8-10.

Nothing like having to work on another sys admin's boxes when
they took the "3" as the max number, when a metadb's disk 
decides to crap out. :-(

My other favorite thing was the sys admins who decided to use
slice 2 as the slice to use in the metadevice.  If you use slice 
2 for this purpose, and that disk fails, when you replace it with 
a disk of the exact same geometry, it thinks it is the same, 
broken, disk.

=Nadine=