Take the 2-minute tour ×
Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

We're experimenting with what VMWare called a "Fully Collapsed DMZ" on blade centre. Basically our DMZ goes straight into a vSwitch and all the security appliances are virtualised.

I've spent days reading up about why this is a good idea and why it's a bad idea, what needs to be done to make it safe, etc, but the one thing I'm having trouble finding is information regarding the best fault tolerance method.

Our edge firewall of choice is pfSense which supports CARP. We've got 10 blades in the cluster, so it's quite feasable to have two or even three pfSense firewalls with VMWare HA enabled and configured internally with CARP that take over eachother in the event of a blade failure. But this seems like a lot of administrative overhead and I'm an un-trusting kind of guy, so it means that I'll be logging into multiple firewalls every week to make sure that all our rules etc have mirrored.

But why bother with CARP when VMWare's FT (even with its single vCPU shortfall) will provide all the features of CARP and as afar as I can tell, less management, stress and concern for my job.

tl;dr:

Is there any compelling reason to use CARP over FT, or vice versa for a software-based firewall?

share|improve this question
    
apart from relying solely on VM security for network security (which is technically off topic)... don't tell publically where this contraption works, security by obscurity might actually help you maintaining this illusion of security. –  knitti Nov 11 '10 at 22:45
    
@knitti - neither of the proposed options involve an "illusion of security". –  EEAA Nov 11 '10 at 22:48
    
@knitti - thanks for your concern, I've read hundreds of pages of whitepapers (they must be serious, they were all in PDF format) of people claiming resolutely that it's either a brilliant idea or a terrible idea. I've come to the conclusion that the only way to know for sure is to try it myself in real life. Our client demands that we permit them to run their own 3rd party security audits on all the systems we deploy for them, so if it fails their audits then I'll look at changing it around. –  Mark Henderson Nov 11 '10 at 22:57

3 Answers 3

up vote 3 down vote accepted

Although I've played with FT for ages now I've yet to find an actual use for it to be honest. Not only is the single vCPU thing a pain but the shear amount of network traffic generated is astonishing. You really do end up using most of a GigE link just for FT, so you end up throwing your vMotion traffic over another vswitch - making the whole thing quite a pain in the backside to be honest. My other concern is that FT only protects you against physical failures, if the FT'ed VM falls over for any reason then you still lose the service as the outage will be perfectly mirrored in the secondary VM.

I'm a cautious guy when it comes to production systems and just don't think FT's worth it right now, hope that changes but I'd sooner have other systems such as clustering/VIPs etc. there instead.

Oh and don't worry about collapsed DMZs, if you're using one of the vShield products I personal think they're as secure as any Cisco box.

share|improve this answer
1  
That's a good point regarding the firewall actually falling over internally. I've seen the amount of FT traffic generated when running some low-utilisation medical software and it's kinda scary, even on a 10Gbps network. Thanks for the advice (as usual) –  Mark Henderson Nov 11 '10 at 23:15

CARP is really designed to let your hosts detect if the other host's network NIC is offline (which is often the case when the physical host is down, but not necessarily)

The advantage of using CARP over VMWare FT would be if VMWare FT behaves differently when the NIC has failed.

If you're comfortable with running your firewall on a VM then the only concern I would have is the FT behaviour on a NIC failure. If a NIC failure does not force the FT to failover, then I would retain CARP.

share|improve this answer
    
Thanks. Our nics are all set up with redundancy at the vsphere level as well. –  Mark Henderson Nov 11 '10 at 23:10
    
Apologies if I wasn't clear, I was directing comment at the VM Guest behaviour during NIC failure. If VMWare Server can detect the Guest NIC Failure, and treat the guest as failed then you have the same behaviour as CARP. –  samt Nov 11 '10 at 23:31
    
The advantage for CARP would be using pfsync (not sure if it's in pfsense) which synchronises the state tables of the firewall, so when the backup firewall is given control, it continues to pass through traffic previously approved by the now failed original primary. –  samt Nov 11 '10 at 23:32
1  
I'd prefer CARP for the reason that the Firewall OS is in "control" of failover. So if you accidentally do something dumb like down an interface, it fails over, where FT would still think the NIC is working (which the physical and virtual hardware is, but the effect would be otherwise). –  Chris S Nov 12 '10 at 3:12
    
Ahh, I follow now. Yeah, that's something I clearly hadn't through through properly. Looks like CARP and regular old High Availability it is. –  Mark Henderson Nov 12 '10 at 5:06

In addition to what Chris S mentioned, which I agree with, I would also go with CARP because what happens when you're upgrading a firewall or doing any other kind of maintenance requiring a reboot or power down? With FT, you're down while it reboots, with CARP, it's completely transparent. Or if you need to change something on the VM that can't be done while it's live, you have to cut off everything. Use CARP and you're in good shape in all those scenarios. FT is mostly to protect from hardware failure, where CARP accommodates any imaginable maintenance needs with no downtime at all, as well as protection from hardware failure.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.