[plug] How-To Murder?

[Craig Ringer]

On Wed, 2004-04-21 at 16:07, Bernd Felsche wrote:

> > I take it that 'doc/install-murder.html' (present in at least 2.2.3)
> > wasn't useful?
> 
> Useful, but a bit spotty in details. I think it assumes that I
> already know a lot more about Cyrus. I don't have a sand-box lab to
> test it so my level of confidence is very low.

That does make a big difference. There's no way you can do a
single-frontend single-backend or two-backend murder test on a couple of
boxes at home, just to get a better feel for the setup?
 
> Add to that the tiny matter of most of the back-end machines being
> deployed more than 3000 km away...

Ack. Not so bad because you won't be upgrading kernels or doing other
low-level tasks, but still it's rately fun to have to work on machines
you can't access physically if you have to.


> > When it comes to authentication, I think most people run murders in
> > combination with distributed authentication schemes like LDAP and
> > Kerberos. I've heard good things about both, especially LDAP with
> > replica LDAP servers on each backend and frontend in the murder.
> 
> SASL is mentioned.

Sorry. Perhaps I should've said that I'd heard good things about
sasl->auxprop->ldap and sasl->auxprop->kerberos, with
sasl->saslauthd->pam->ldap (what I currently use) also working quite
well.


> Erm no. ISC DHCP first pings the address before handing it out.
> The extra delay is a worthwhile sacrifice compared to breaking a
> network connection to an existing host; which with Windows can
> easily result in file corruption.

I'm glad to hear that, having never noticed it in the ISC DHCPd docs. It
seems rather sensible, really.

Craig Ringer