Re: connectivity redundancy setup question

From: Menno Duursma (menno_at_desktop.lan)
Date: 11/19/03 

Date: Wed, 19 Nov 2003 09:56:42 GMT

On Wed, 19 Nov 2003 05:38:17 +0000, Noi wrote:
> On Tue, 18 Nov 2003 18:20:53 +0000, Menno Duursma without thinking wrote:
>> On Tue, 18 Nov 2003 17:37:58 +0000, Noi wrote:
>>> On Tue, 18 Nov 2003 14:36:54 +1100, NS without thinking wrote:

[ Snip. ]

>> Problem i see with any of these setups is: there will still be a single
>> point of failure - in the main box itself. (And if you can live with
>> that, i don't see the point in useing 2 boxen).
>>
> The issue wasn't if the box fails just the connection.

In that case i don't realy see the point in maintaining more then one box
for this.

> Avoiding failure because the box fails pushes the redundancy further down
> the network.

Yes. however some device without a HD (say a switch) is far less likely to
fail then a PC. If your concerned about that failing, you'd need at least
two and have them do spanning-tree (connect both boxen to both swithes).

Non of this would make any sence if you don't use UPS (or no brake) power
suply also. And connect the consoles to a terminal-server as well.

> Assuming all connected boxes route traffic out the eth0 dev an external
> dev (router/switch) would need to switch traffic from the failed box to
> the redundant box or manual intervention to switch plugs.

Nope. All you need is an alias IP adress - which only one box binds to
it's MAC - upon fail of the main box: the backup machine aliases the
shared IP to it's MAC and uses something like "arping" to update the
clients ARP caches. (falling back would be the other way around.)

Or, you could use some kind of hearbeat over a second conection (which
might be ethernet, serial, paralel, USB, or whatever.) And take over the
MAC adress as well - using "ifconfig hw ether". I which case there
wouldn't be any need to update any ARP caches.

>> To avoid that, you could set the main machine to "ifconfig down" it's
>> internal (alias) interface and have the backup machine take it over
>
> That would kill internal access to that server.

Nope, it's an alias (eth0:0 or something). Thus you can still access it
via the adress bind to eth0.

>> (upon getting packet-loss) dail-out and mail you "where on backup now"
>> :-). In this setup it might be a good idee to have at least two
>> connections between main and backup machines (as else the one connection
>> might fail, triggering the backup routine).
>
> I agree? For 2 box solution it's tricky.

Yes, i'd prolly stick two NICs (for "bonding") and a Flash HD in one
box and connect both the broadband and POTS modems to it. With a simple
UPS for the power supply.

> A script on the dialup box
> continually tests connectivity to the broadband box. If that connection
> fails the script changes the IP address of the dialup box to that of the
> broadband box and initiate the dialup connection.

You'd still need a way to tell the clients, the IP is to be found on some
other MAC (Media Access Controller) in the network.

> When broadband box comes
> back online the script brings down the dialup connection, resets the
> ipaddr and after x minutes continuous connection to the broadband box.
>
> Of course a similiar script on the broadband box to change and reset
> ipaddr. Something like that would work in either the broadband down but
> broadband box up or the broadband box down situations.

This is more or less what i was rambling on about... 

-- 
-Menno.

Relevant Pages

  • Re: DC Redundancy Not Working...?
    ... with no FAIL or WARN messages. ... Starting test: OutboundSecureChannels ... The connection was aborted by the remote WINS. ...
    (microsoft.public.windows.server.general)
  • Re: Problem with gethostbyname
    ... WiFi) then the program exits which is undesirable. ... while and then tries again until the connection is available again. ... once a call has failed all subsequent calls fail until the ... I would be very surprised if it gethostbyname() that is causing the program ...
    (uk.comp.os.linux)
  • Re: How can I tell when a remote TCP connection is closed?
    ... recv won't fail - it will return zero bytes read for graceful ... Microsoft MVP, MCSD ... >> I have a client with a TCP connection to a server. ... >> The remote end is doing a shutdown and close on the socket. ...
    (microsoft.public.win32.programmer.networks)
  • Re: Broadband internet connection
    ... connection with the same UID and password - again it fails because of UID ... Pasting user ID's and passwords can sometimes fail. ... box, login would fail. ...
    (microsoft.public.windowsxp.network_web)
  • Re: Unipod Network Problem
    ... The router is not just ... First, I do take your points, Chris - and the points ... I suppose when using BT's latest broadband hub 'phone ... broadband connection.. ...
    (comp.sys.acorn.networking)