Re: LTSP Bonding / Load Balancing
On Mon, Aug 31, 2009 at 02:19:20PM +0200, RalfGesellensetter wrote:
> just arrived has a new switch with 2 Gigabit Uplink (yeah!).
> We discussed about bonding and bottle necks in bandwidth or CPU.
>
> As I have a 2nd (identical) LTSP machine, why not do something like
> "server bonding" - or even load balancing.
>
> http://wiki.debian.org/DebianEdu/Documentation/Etch/HowTo/NetworkClients
>
> describes how to do this by moving thin clients to 10.0.2.0/23.
>
> Iff the switch has v-lan, we could just split the room (14 clients) to
> make 7 client each use one LTSP server.
>
> Otherwise it should be easy to make those clients see both LTSP,
> provided that
> - the servers get different IPs (192.168.0.254 + 253)
> - the less busy server is faster with DHCP
that last part is unlikely to really happen. for whatever reason, i've seen
very loaded servers that were quick to respond to DHCP, and underloaded ones
that were slow. that method does at least provide a cold failover; if one
server is down, reboot the affected thin clients and they'll all still work.
there is infrastructure in LDM to handle load balancing at boot time, which
isn't ideal as the server load may change by the time the user logs in, but
simple round-robin approaches are very doable.
in the not yet released version of LDM, it should be possible to test for
load-balancing at login time, which makes more sense.
live well,
vagrant
Reply to: