On Wed, Sep 25, 2002 at 01:53:06PM +0200, Ragnar Wisl�ff wrote:
> Sitat Hans Ekbrand <[EMAIL PROTECTED]>:
> 
> 
> > > 
> > > rsync from cron.
> > 
> > But what if a user logs out, and in again before the cron job is
> > run?
> > 
> 
> If you have load balancing rather than fail-over then you risk 
> that such a user ends up on the node that still has not been 
> synced with the contents of the other.

Indeed. And the solution I headed for combined load-balancing and
fail-over.

> For a fail-over solution, 
> the chances of someone logging out and then in again between 
> syncs - and the passive node having taken over from the active - 
> should be very small indeed.

Yes, that was not the problem I thought of.

> Anyway, rsync is smart enough to 
> only update what's changed in individual files, so the problem is 
> only transient. Unless you use shared storage it very difficult 
> to totally guard against data loss/corruption.

Well, that is what I am trying to solve here.

> Depending on load and capacity of servers and interlinking (I'd 
> say three NICs in such boxes - the additional for the syncing) 
> you should be able to have a frequency of 3 - 5 minutes between 
> syncs. Even with rsync running via ssh (and thus loading the 
> machines heavily) you can move a lot of data in 3 minutes on a 
> dedicated 100 Mb/s link. 

IMHO, a setup that only sync when it is needed (once per login/logout)
is the only thing I would be satisfied with. Even though rsync is
smart, it takes alot of CPU, and to run it frequent is a waste of CPU.

If users seldom logout, a nightly rsync would be necessary of course.

Also note that loadbalancing with double dhcp-servers might not work
as you want it to, if the thin clients never reboot.

-- 

Hans Ekbrand

Attachment: msg08151/pgp00000.pgp
Description: PGP signature

Reply via email to