[Linuxha-users] Update on Linuxha.net 0.8.3 progress

[Simon Edwards]

Michael,
	Sounds like a similar environment to the hardware I use! In short, yes,
the new networking code can easily handle this type of scenario.

Regards,
Simon.

On Mon, 2005-02-07 at 10:14 +1000, Michael Mansour wrote:
> Hi Simon,
> 
> Change is never a bad thing especially when there's much benefit to it. I'm 
> starting the build of my production cluster this week, and I was conerned that 
> with v0.8.2 the networking wouldn't have worked as well as I wanted it to. 
> I'll explain as follows.
> 
> I want 3 network cards in each machine, 2 for the "client" network, and 1 for 
> the "drbd" network. Each network card I have so far is 10/100Mbps running at 
> 100Mbps full duplex. However, to allow for fast transmission of the drbd 
> packets and writes to the standby node, I have purchased 2x 1Gbps network 
> cards to replace the 100Mbps FD cards for the drbd network. As I have no need 
> to go to the expense of purchasing a gigabit switch, I decided to just use a 
> cross-over cable between the 1gbps cards. At this point, I started to think 
> how I would allow traffic to flow through between the drbd network, and I 
> believe the best way to do this would be to have that in a separate subnet to 
> the main network.
> 
> I believe the changes you have made in 0.8.3 will allow this yes?
> 
> Michael.
> 
> > Hello all,
> > 	Just a quick update on progress of the Linuxha.net 0.8.3 release. I 
> > am hoping this will be released in the next day or so. However 
> > rather than the online file system size change support 0.8.3 will be 
> > shipped with a completely new networking sub-system. I deemed these 
> > changes necessary due to several limitations of the previous code, 
> including;
> > 
> > [1] No clear division between cluster and application configuration
> > files for the cluster network topology settings.
> > 
> > [2] The fail-over response was taking too long (up to 3 seconds).
> > 
> > [3] The code made use of multiple IP addresses on the same subnet -
> > which is not considered a good thing.
> > 
> > [4] The application was only able to apply IP addresses on one card and
> > on one subnet on that card.
> > 
> > The new code addresses all these problems and overall results in
> > improved fail-over times, reduced code size and a more logical
> > separation between application and cluster configuration information.
> > 
> > These changes are handled by having a separate network monitoring daemon
> > which is always running when the main cluster daemon runs on a node.
> > 
> > How does this all affect existing applications? Well changes with
> > existing configurations of both the cluster and application
> > configuration files will be necessary - but the changes are not huge,
> > for example in the cluster configuration file you now define "networks"
> > - groups of cards that can provide IP addresses on the same subnet, i.e;
> > 
> >         <node>
> >                 <name>server1.linuxha.net</name>
> >                 <network name="main" cards="eth0,eth1"/>
> >                 <network name="drbd" cards="eth2"/>
> >         </node>
> >         <node>
> >                 <name>server2.linuxha.net</name>
> >                 <network name="main" cards="eth0,eth1"/>
> >                 <network name="drbd" cards="eth2"/>
> >         </node>
> > 
> > It is possible to define 0,1 or more networks and call them whatever 
> > you wish (here called "main" and "drbd").
> > 
> > In the application configuration file the network section might look
> > like the following;
> > 
> >         <networks>
> >                 <network net="main" ip="192.168.0.102,103"
> >                         checklist="192.168.0.1" checkpercent="100"
> >                         pingtype="icmp" pingtimeout="3"/>
> >                 <network net="drbd" ip="192.168.100.12"/>
> >         </networks>
> > 
> > In the above example the application hosted two IP addresses on the
> > "main" network and actually one on the "drbd" network. Now an
> > application can defined 0, 1 or more IP addresses - and these can be 
> > on multiple networks as well. Notice that you do not define the 
> > cards that can be used - all topology information is now handled at 
> > the cluster level to keep the application configuration as 
> > straightforward as possible.
> > 
> > The above is a sample of the changes - one or two other changes will 
> > be necessary but shortly after the code is released I'll provide suitable
> > documentation to migrate.
> > 
> > Although a big change I believed it was necessary to ensure that the
> > version 1.0.0 reaches the quality I wish it to.
> > 
> > Regards,
> > Simon.
> > 
> > _______________________________________________
> > Linuxha-users mailing list
> > Linuxha-users at linuxha.net
> > http://linuxha.net/mailman/listinfo/linuxha-users_linuxha.net
> ------- End of Original Message -------
> 
> 
>