Score:1

What would be the benefit of using a Stream Cipher which achieves encryption performance of 0.01 cpb using 1 CPU thread?

cn flag

If such throughput could be achieved what would the benefits for a data center or IoT devices in terms of power saving. Also I was wondering if such speed was possible to be realized by a cipher, would that dramatically change cryptography in terms of encryption/decryption speed? If that is possible in the future, would that allow to significantly upgrade transmission and many other things on the internet?

enter image description here

As an example in this paper of Intel(see page 5) using AES acceleration and optimazations they managed to achieve encryption speed using AES 256 CBC version of 0.49 cpb. Which is great by the way!

However, if it was possible to reach for example 0.01 cpb just by using 1 thread, instead of 12 like as shown in the screenshot. How important this would be for cryptography in general?

I hope my question makes sense!

Score:5
fr flag

In most cases, our cryptography is already faster than we need it to be. For example, on my 3-year old laptop with a 1.9 GHz CPU, I can use AES-GCM with a 128-bit key at 5.8 GB/s and with a 256-bit key at 4.5 GB/s, all on a single thread. Those speeds are sufficient to saturate a 10 Gb/s link and the former is fast enough to saturate a 40 Gb/s link, neither of which I have. On our servers at work which have much faster connections, we also have CPUs that are much faster, and encryption (e.g., SSH) is also not the bottleneck there.

Even on very low end devices without hardware acceleration, ChaCha can perform extremely well. It can be implemented with great performance on anything which supports a 128-bit SIMD implementation, and even performs well on systems without SIMD units at all.

In my experience, it is much harder to get people to use a secure design, algorithm, or protocol than it is to get cryptography which performs well. People still insist on using insecure algorithms because they are fast without realizing that there are better, secure alternatives which meet their needs and may even perform better. Thus, I find it unlikely that an extremely fast design would revolutionize cryptography because the problems I tend to see lie elsewhere.

Tunnel_Vision avatar
cn flag
Great answer! Thank you very much for your input. I see your points and I understand. However, I keep thinking, are there areas that such speeds could come in handy? Maybe for encrypted data that are being transmitted from space? Or from data centers and companies that are analyzing and using huge amounts of data? Could there be some applications where achieving high throughput like this, while having a sufficiently secure encryption algorithm could be needed?
bk2204 avatar
fr flag
The problem is that you effectively need some way to ingest that much data (e.g., a network card) and usually we can already cope. If you really need better performance, you can just add an accelerator card with a chip that has a bunch of cores processing data in parallel. Many Wi-Fi cards already handle the encryption instead of the CPU, so this kind of offloading isn't unknown. But in probably over 98% of cases, we already perform well beyond expectations and our symmetric crypto isn't a concern.
Tunnel_Vision avatar
cn flag
I agree with you that we perform well beyond expectations due to the usage of specialized hardware. When I was serving in the army, senior officers were using encrypted phones with all the necessary chip performance enhancements. Yet, it was a very expensive solution. Also, what about all the extra hardware that is being installed to data centers every now and then for the exact purpose of higher throughput. Yeah, It isn't a concern, but my point was if we could somehow invent a better algorithm, (software solution) wouldn't we be able to lower the cost significantly?
mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.