topic Re: High availability end-to-end in Operating System - HP-UX

High availability end-to-end

Stexxe — Fri, 23 May 2008 09:17:56 GMT

I have
4 physical ethernet adapter: lan1, lan2, lan3, lan4 that should e dedicated di backup network
3 virtual machines (HP UX 11 v2) that needs to backup their data
2 external switches (Cisco)

And I want to create a full redundant solution...

If I create a single vswitch that will manage all the 4 cards (configured in APA-link aggregation) I will have 2 SPOF: the vswitch and the single switch where I'm forced to connect all the 4 pnic

Is there a way to create:
vswitch-A: actively connected to lan1, la2; passively connected to lan3, lan4
vswitch-B: actively connected to lan3, la4; passively connected to lan1, lan2

that works like a cluster considering that:
1- I want to use all (at least the most...) the bandwidth of the four adapters
2- I want to split the four adapters to the two switches for HA
3- in case of foult the system should react (hopefully) automatically

Thanks in advance,
Stexxe

Re: High availability end-to-end

Torsten. — Fri, 23 May 2008 09:24:13 GMT

IMHO the vswitch in HPVM is not a SPOF, because a virtual device will not fail.

I think you should consider to adjust your APA config like the Hot Standby Configuration for High Availability or failover mentioned in the APA guide.

Re: High availability end-to-end

Stexxe — Fri, 23 May 2008 10:55:44 GMT

Thanks Torsten for your reply, but let's supposing that the vswitch will never fail, in any case I don't want to configure the 4 pNIC in failover becouse in this case I will not use all the bandwidth.

What's about this configuration?

On each VM (VM-A, VM-B, VM-C, VM-D) configure 2 vNIC (Backup0 and Backup1) in a Failover group (so active-passive)

On the VM-A
The first vNIC(active) is connected to the vswitchA (connected only to the lan1 and lan2 in APA-link aggregate)
The second vNIC is connected to the vswitchB (connected only to the lan3 and lan4 in APA-link aggregate)

On the VM-B
The first vNIC(active) is connected to the vswitchB (connected only to the lan3 and lan4 in APA-link aggregate)
The second vNIC is connected to the vswitchA (connected only to the lan1 and lan2 in APA-link aggregate)

and so on for the other VM...

Re: High availability end-to-end

Torsten. — Fri, 23 May 2008 10:57:56 GMT

From the manual:

Do not run the following types of applications on a guest:

...

Auto port aggregation (APA)

Re: High availability end-to-end

Stexxe — Fri, 23 May 2008 11:13:47 GMT

Ok, so are there any other options (avoiding installation of service guard on guest...) ?

Re: High availability end-to-end

Torsten. — Fri, 23 May 2008 11:21:29 GMT

How about adding another 4 port NIC for redundancy?

Re: High availability end-to-end

Stexxe — Fri, 23 May 2008 11:24:34 GMT

But how can I use also the bandwidth offered from these additional ports?

I don't want to use an active-passive configuration...

Re: High availability end-to-end

Duncan Edmonstone — Fri, 23 May 2008 14:39:51 GMT

Hi,

When you mention 4 x NICs are these 4 x GbE or 4 x 100BaseT?

Because if you're talking about GbE, I'd consider whether something that needs a combined throughput of 4Gb/s is appropriate for a VM anyway.

Remember that with virtualised IO, all that IO load will also load up your CPUs as well. The situation is much improved with AVIO in VM3.5, but you still need to consider it.

Why do you think you need 4Gb/s for your backup network - do you have a lot of data to backup in a short window? There are other ways of acheiving this. Have you read the WP on backup strategies in VMs?

http://docs.hp.com/en/13163/BackupStrategiesforIntegrityVM.pdf

HTH

Duncan

Re: High availability end-to-end

Stexxe — Fri, 23 May 2008 14:54:18 GMT

Let me say that we are talking about 100BaseT adapter becouse my real problem are not the bandwidth but how can I achieve the high availability...

Actually if I configure the 4 cards in APA (link aggregation) and I connect the physical card to a single switch (I'm forced to do this in order to use the link aggregation protocol) and this fail I will lost all the connection to the VM.

My "wish" is to create 2 APA groups:
APA1: lan1, lan2 (connecter to PhysicalSwitchA)
APA2: lan3, lan4 (connecter to PhysicalSwitchB)

Create 2 vswitchs:
vswitch-A connected to APA1
vswitch-B connected to APA1

On the VM create 2 vNIC each of them connected to a different vswitch.

In this case if I have a failure of PhysicalSwitchA I will lost the lan1, lan2, vswitch-A but the VM will be able to use the vswitch-B.

I've read something about nested APA, but i haven't found any additional material about... could be a solution?

Bye,
Stexxe

Re: High availability end-to-end

Duncan Edmonstone — Fri, 23 May 2008 15:04:36 GMT

It's certainly possible to create a failover group out of link aggregates if thats what you mean by 'nested APA'. See this section of the APA manual:

http://docs.hp.com/en/J4240-90039/ch03s01.html#c1067293

HTH

Duncan

Re: High availability end-to-end

Stexxe — Fri, 23 May 2008 15:18:24 GMT

For "nested APA" I mean:

"A nested APA device is a grouping of other APA devices. For example, if lan900 is an APA device composed of physical network interface cards (NICs) lan1 and lan2, and lan901 is an APA device composed of physical NICs lan3 and lan4, the HP network configuration utilities allow you to define a nested APA lan902 device that is composed of lan900 and lan901."

http://docs.hp.com/en/T8671-90002/ch04s06.html

With the "Sample Failover Group Using Link Aggregates Configuration" the stand-by card is unused... ^_^

What will happen istead with the Nested APA?

Bye,
Stexxe

Re: High availability end-to-end

Duncan Edmonstone — Fri, 23 May 2008 15:39:00 GMT

This is just terminology for the same thing. What I pointed you to in the manual is the same thing as is referred to as nested APA in the VM manual.

In a link aggregate that is configured for load balancing rather than failover, *all* the conenctions have to be to the same switch, so you *never* get HA when load balancing.

HTH

Duncan