I have the last half year build serveral Skype for Business solutions where we have implemented Enterprise Edition.
Based on what several MVP and Microsoft best practise they recommend DNS LoadBalancing.
When reading on TechNET this is one of the article about Loadbalancing
As mentioned in several bloggs and on Technet, well all say that searching for error is more easy with DNS LBA. Well that can be true because you don't have to search in HLB vml og physcal unit. But then you buy a HLB without a service and support agreement? And if it's critical you buy a 24/7/365 4 hours responce time agreement.
If you look at downtime based on how client works and what is said around. The Skype for Business client shall do it better than older clients. But is this the fact?
From my experience I have seen on Lync 2010 with Lync 2010 client, that for internal and external users that downtime when take down on frontend server or edge server in the pool can give up to 90 seconds downtime.
I have seen on Skype for Business running DNS LBA the same result where all the users are external that downtime on DNS LBA has been from 7 to 90 seconds during meeting, where Powerpoint, audio and video are running.
What make this also interesting is that if you run a design over the Skype for Business planning tool, and you fill out that 100% of the users are external, well then it recommend that you should use Hardware Loadbalancing.
But running HardWare LBA the same downtime is just 7 seconds no matter what service you tries to to stop.
So why use DNS LBA vs HLB? Well from Microsoft and I think that is why the MVP say it also is that DNS LBA is more easy to configure and manage, and the cost of the Hardware Loadbalacer is an issue.
I have asked also the same question in techcommuity and still I have no comment on this but only my new question on it. See this link:
But when it comes to security and downtime, will this cost for a customer implementing Skype for Business OnPrem, where they use 2000 USD more matter? Well from my point, NO. All that matters is that downtime is low as possible.