[Openais] Openais Digest, Vol 90, Issue 1

[Darren Thompson]

mumtaz

I'm still not convinced that your use of corosync ring redundancy is even
solving the correct problem in your case, it looks to me that you have an
invalid network configuration with the two interfaces on the same subnet,
that may be the root of your problems.(You have two interfaces in the same
lan, each with separate IP addresses... I'm not sure that is even good
practice).

I'm not sure why you say this: "Also, bonding of interfaces does not work
for me as I need to interfaces each with a separate address." as I have
regularly used exactly that configuration without error for the last two or
so years...

If you want to separate the Heatrtbeat traffic from other IO traffic you
could just setup VLAN interfaces over the top of the bond.

In either case if you use 802.3ad mode it gives you almost twice the
bandwidth per host, so you get fault tolerance and more bandwidth...
win/win.

Try it, you may be surprised...

Regards
Darren


On 26 January 2012 10:07, M Siddiqui <msiddiqui at live.com.pk> wrote:

>
>> Date: Wed, 25 Jan 2012 14:02:39 +1100
>> From: Tim Serong <tserong at suse.com>
>> To: openais at lists.linux-foundation.org, discuss at corosync.org
>> Subject: Re: [Openais] HA Cluster Connected over VPN
>> Message-ID: <4F1F70CF.6030705 at suse.com>
>> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>>
>> On 01/25/2012 12:32 PM, M Siddiqui wrote:
>> > Hi there,
>> >
>> > I have a situation where two cluster nodes are connected over the VPN;
>> > each node
>> > is configured with two interfaces to provide ring redundancy for
>> corosync:
>> >
>> > NODE1:
>> >    eth1: 192.168.1.111/24 <http://192.168.1.111/24>
>> >    eth2: 192.168.1.112/24 <http://192.168.1.112/24>
>> >
>> > NODE2:
>> >    eth1: 192.168.1.113/24 <http://192.168.1.113/24>
>> >    eth2: 192.168.1.114/24 <http://192.168.1.114/24>
>> >
>> > corosync version 1.4.2
>> > transport udpu (multicast has the same issue)
>> >
>> > Since two nodes are geographically distributed and connected over the
>> VPN,
>> > configuring each interface in a different subnet is not an option here.
>> >
>> > Now corosync got confused due to same subnet; how we can handle this
>> > situation?
>> > What is the experts recommendation? Thanks in advance for the answer.
>>
>> I'm pretty sure if you're doing multiple rings, they need to be on
>> separate subnets.  Question: if you're going over a single openVPN
>> instance, you only really have one communication path between the nodes,
>> right?  In which case, redundant rings won't actually help.
>>
>
> I see. Thanks!
>
> Actually in my setup I am using two interfaces on each node:
> eth1 for heartbeat and eth2 for some data aggregation from other
> hosts on the same network as well as hosts across the VPN.
>
> Now I agree there in one communication path for hosts across the
> VPN but we can avoid congestion while aggregating data from hosts
> on the same network; (I mean all host on one end of VPN). In this
> situation, even if we don't configure eth2 as a backup ring in
> corosync.conf
> still corosync got confused and does not work.
>
> Also, bonding of interfaces does not work for me as I need to interfaces
> each with a separate address.
>
> regards,
> mumtaz
>
>
>> Also, you probably want the discuss at corosync.org list.
>> openais at lists.linux-foundation.org is deprecated, for lack of a better
>> term.
>>
>> Regards,
>>
>> Tim
>> --
>> Tim Serong
>> Senior Clustering Engineer
>> SUSE
>> tserong at suse.com
>>
>>
> _______________________________________________
> Openais mailing list
> Openais at lists.linux-foundation.org
> https://lists.linuxfoundation.org/mailman/listinfo/openais
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/openais/attachments/20120127/d65c4762/attachment.html>