Server Failover

macadam · 05-16-2007, 04:59 AM

Hi!

I am running 2 servers.
I use round robin to balance the load but that's not what I really want as one server is 2 times slower than the other.
What I would like is to use the fast server as main server and the slow one as backup.

I need thus to set up a failover system.

What is the easiest way to do that?

Regards,

macadam

acid_kewpie · 05-16-2007, 06:27 AM

depends at what level you want to implement is, but essentially you can use a heartbeat to pull an ip address across if it stops responding in the slave's opinion. check http://linux-ha.org or i was looking at a very simple specific implementation here... http://www.onlamp.com/pub/a/onlamp/2...inuxhacks.html

macadam · 05-16-2007, 06:39 AM

Hi,

I have just written a script that pings the main server and updates the DNS zone file accordingly.
Should work as I have a very short TTL.

Regards,

christophe

acid_kewpie · 05-16-2007, 07:23 AM

can't say i like that approach. if this is at all internet facing, you can't guarentee those TTL's are adhered to at all. seems like a long long way round if those two boxes literally are next to one another...

macadam · 05-22-2007, 03:10 AM

Well the boxes are 1200km from each other.
That's the problem. Otherwise the system would be much more easier.

acid_kewpie · 05-22-2007, 03:45 AM

and you never thought that might be useful to say that before?

macadam · 05-22-2007, 06:10 AM

Indeed it might have been wise to mention it.
But for me it was obvious...why should I have 2 servers at the same location.
In this case, we can only cover a hardware failure...not a network or power failure.

Sorry not to have described the problem more accurately.

p_s_shah · 05-24-2007, 03:17 AM

Hi,

Would you please give us idea, how you are load balancing two servers using round robin ?

Do you have any other scheduling option other than round robin, so that out of three requests, two requests will be server by main server and one by second server ?

acid_kewpie · 05-24-2007, 03:27 AM

Quote:

Originally Posted by macadam

Indeed it might have been wise to mention it.
But for me it was obvious...why should I have 2 servers at the same location.
In this case, we can only cover a hardware failure...not a network or power failure.

Sorry not to have described the problem more accurately.

obvious? so you know nothing about conventional resilient architectures then...? two servers, two electrical supplies backed up by two ups's two halfs of a resileint network, two internet feeds... one site.

acid_kewpie · 05-24-2007, 03:27 AM

Quote:

Originally Posted by p_s_shah

Hi,

Would you please give us idea, how you are load balancing two servers using round robin ?

Do you have any other scheduling option other than round robin, so that out of three requests, two requests will be server by main server and one by second server ?

that'll be DNS, and not likely to help you one bit. DNS RR's only help spread the load of a busy server with a lot of clients. if it's only a few then it'll most likely be very unpredictable and asymmetric. may well be possible to just list one servers A record twice i guess.

macadam · 05-27-2007, 02:42 AM

Quote:

Originally Posted by acid_kewpie

obvious? so you know nothing about conventional resilient architectures then...? two servers, two electrical supplies backed up by two ups's two halfs of a resileint network, two internet feeds... one site.

Sorry chris but by experience:

- even a good hosting company can have all their connections down even with redundancy like you mention above (I have had the case with 3 providers yet)
- it is recommended to have 2 separate class C networks (at least for DNS)...difficult to find a provider having those class C's.
- so except for the power supply...well I need to select different hosting solutions at different places...