Michael Tremer

Is AES-NI a requirement for everyone?
by Michael Tremer, May 2

Yesterday, I read that pfSense won’t support systems that do not have AES-NI from version 2.5 onwards any more. They state, that systems with AES-NI perform better and that this is available in many Intel processors from 2008 and AMD processors from 2010.

I am a little bit surprised by this post when it scrolled along my Twitter timeline and so far I cannot make any sense out of it. Here is why:

AES-NI does not mean performance

In some earlier and related posts on the planet, I have been trying to illustrate this point – so I will keep it short here.

We have come across a number of systems that have AES-NI, but they are overall not faster than a system that does not have these instructions. Our wiki even has a short table with some of these examples. AES throughput rather depends on clock speed of the processor and that can make up for missing AES-NI.

Performance isn’t everything

I run many IPFire installations that are out somewhere in the field and running on rather compact and efficient hardware. They work in an IoT scenario and transfer only a few kilobytes up to maybe a megabyte of traffic a day. That data is going through a VPN, but it does not matter at all if the system has AES-NI to encrypt it or not. It is effectively as fast either way.

However, if VPN performance is an issue, because you are running IPFire in a datacenter and need to have a VPN connection that can transfer up to 10G, then you will need that extra power. But this is not a scenario that everyone who is using IPFire has.

~14% of all IPFire systems actually have AES-NI

As of today and according to fireinfo (which is probably not representative, but a the best source that we have), only about 14% of all known IPFire systems have AES-NI. That’s not a lot. It does not always come with all Intel or AMD processors and it is not necessary. It appears that the majority of IPFire users doesn’t need AES-NI.

Security

For the security of a VPN or where ever AES is used, it does not matter if AES-NI is used or if AES is performed in software (assuming that no implementation has any implementation problems).

Conclusion

So in conclusion, I do not have any technical reasons left over that I could consider. Those I listed above to not justify to me in any way that AES-NI is a must and that people need to have hardware that has it. Some of them even proof the opposite.

And although I also have been complaining a lot about that we have to support a lot of old and outdated hardware – and by that I mean decades old – this seems to be the only reason left: To incentivise people to buy new hardware which is quite conveniently only a click away.

And again, we do that, too. Having some kind of standard hardware makes things easier. However, that does not always make sense for everyone and therefore we leave that decision to you and work on making IPFire run on as many platforms as we possibly can.

And to make that crystal clear: Systems without AES-NI will be supported just as well as those that do have AES-NI.


Posted: May 2 • 2137 views