Re: A Few Questions

["Mark H. Wood" <mwood@IUPUI.Edu>]

On Fri, 5 Apr 2002, Geoff Silver wrote:
> > I think you need to look for some load-balanceing or clustering solution
> > (Turbo Linux?) to do this.  slapd as far as I know does not have a "fail
> > over" method,  which is what it sound like you are actually looking for.
>
> Wow, I'm surprised there isn't the ability to do this.  So slaves are only
> really good for load balancing, not really redundancy (except manual
> failover).

What happens when you simply insert A records for the master and the
backup, in order, and do *not* enable round-robin responses?  The master
should handle the bulk of the work, and if it's fast enough it should be
able to completely starve the backup, no?

> Hmmm.  I wonder if libnss-ldap, libpam-ldap, etc contain (or could
> contain) code to fail over to a backup if the master is down.  Not quite
> as efficient, since the list of backups would have to be given to every
> application, but would still provide for some redundancy.
>
> Has anyone considered adding this sort of functionality (is it even in the
> protocol specs?)  Since the master knows what slaves exist, it would seem
> to make sense that a client could cache a list of slaves from the master.
> This sort of redundancy would seem to be a killer in a large scale
> corporate environment.  Just a thought.

I'd be interested to know why this is so much better than, say, load
balancing across N+1 servers (that is, you size the server set so that you
can lose one and still handle the maximum expected load without
significant delay).  Having a resource which does absolutely nothing
except in an emergency makes sense if it is, say, a fire extinguisher
(which has no non-emergency function) but I don't see the sense in having
redundant equipment stand idle when it could be continuously tested by
sharing the load.

-- 
Mark H. Wood, Lead System Programmer   mwood@IUPUI.Edu
MS Windows *is* user-friendly, but only for certain values of "user".