Re: Oracle 9i RAC vs Hardware clustering (like HACMP)

Praveen wrote:

> Hello,
>
> We are exploring the options of high availability. Harware clustering
> seems to be the first choice and using Oracle RAC on a hardware
> cluster would further enhance availability. But could anyone let me
> know the advantages of using Oracle RAC instead of plain hardware
> clustering such as HACMP.
>
> Thanks,
> Praveen

There's been a lot of this lately. It's what happens when you don't keep in mind the brutally simple distinction: Hardware clustering involves hardware; RAC involves instances.

So, what does RAC give you that clustering doesn't, at least in the high availability arena? It gives you failover of instances, where hardware clustering gives you failover of hardware.

Specifically, it will be quicker to flip to an instance which is already running (and which has already begun instance recovery of the failed instance) than to get your hardware cluster to start a new instance from scratch. The difference could be between four and five nines -that is, a hardware cluster might give you 99.99% availability. The speedier failover to an already-running instance that RAC gives you might yield a 99.999% availability. Or at least, that *sort* of difference -perhaps even a rather larger one.

Depending on circumstances, the differences could sometimes be huge. In a hardware failover scenario, the listener on the second node has to be started from scratch, and nobody can attempt to re-connect until it has been. Immediately after that, you can confidently expect it to be swamped with connection requests, as every user simultaneously demands failover to the new instance, and queues to be processed accordingly. In RAC, because the second instance was always there, always running, it is quite possible to arrange for every user to have a 'shadow' connection to the second instance, so that at failover time, all they have to do is to activate a connection that already exists. The Listener doesn't sweat anywhere near as much. (It's called Transparent Application Failover (TAF), and specifically refers to the PRECONNECT connection method).

And what hardware cluster is going to be able to resume that 4 hour report that failed midway through from the exact point it got to at the point of failure? But with RAC, and with TAF, that's exactly what can happen (quite a big plus if you're a warehouse that runs gigantic reports).

And so on... ultimately, asking what RAC has over hardware clustering is a bit like asking what an airplane has over the automobile. Or what chalk has over cheese.

Regards
HJR Received on Mon Oct 18 2004 - 14:13:06 CDT