Coda File System

Re: replicated volume that is not on the SCM

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Fri, 10 Nov 2000 17:01:24 -0500
On Fri, Nov 10, 2000 at 01:38:19PM -0800, Josh Logan wrote:
> 
> I have three machines.  Is it possible to have the 2nd and 3rd machine
> share a volume that is not on the SCM?
> 
> I would like to have a central SCM machines that shares it's contents to a
> bunch of remote machines, but I want a pair of remote machines to have
> local content that is not pushed back up to the SCM.  Is this possible, or
> does the SCM always have to be part of a volume?

No, the SCM is only special in that it is the read/write copy of the
shared databases. Users, passwords and volumes creation is done from the
SCM, although the volumes can reside on any VSG (volume storage group).

The VSG's are defined in /vice/db/VSGDB, when it contains f.i.

    E0000500 notscm1 notscm2

# createvol_rep newvolume E0000500

will create the replicated volume on the servers notscm1 and notscm2.

> 
> BTW, this currently looks like:
> root_at_sysops1:~# ls -l /coda/
> total 5
> lrw-r--r--    1 joshl    nogroup         7 Nov 10 13:25 local -> #local
> lrw-r--r--    1 joshl    nogroup        11 Nov 10 13:26 local-stu ->
> #local-stu
> drw-r--r--    3 root     nogroup      2048 Nov  9 10:57 office
> drwxrwxrwx    2 root     nogroup      2048 Nov  8 16:28 staff
> 
> V_BindToServer: binding to host joshl-laptop.internap.com
> Wlocal.1 I3000006 H3 P/usr/opt/CODA/vicepa m0 M0 U2 W3000006 C3a0c6716
> D3a0c6716 B0 A0
> Wlocal-stu.1 I3000007 H3 P/usr/opt/CODA/vicepa m0 M0 U2 W3000007 C3a0c675a
> D3a0c675a B0 A0
> GetVolumeList finished successfully

Maybe the VRDB and VLDB databases haven't been replicated to all servers
yet? Check the timestamps/sizes of /vice/db/VRDB and VLDB on the
servers.

> Also it looks like the max servers a volume can be on is 8:
> codaroot 7f000000 3 1000001 2000001 3000001 0 0 0 0 0 E000010
> 
> Is that a max, or just placeholders.

It is a max, and because Coda uses write-all,read-one replication it is
probably not that useful either to have more. Double or triple
replication is in most cases sufficient.

> I'm sure reintegration gets harder with more servers, but we may want more
> than 8 machines as we deploy into more and more offices...
> 
> 							Later, JOSH

Spread volumes across different VSG's,

E0000100 srv1 srv2 srv3
E0000101 srv4 srv5 srv6
E0000102 srv7 srv8 srv9
E0000103 srv10 srv11 srv12

See, 12 servers deployed with triply replicated volumes.

Jan
Received on 2000-11-10 17:15:55