Linux kernel RNG enhancements for 5.19 | Hacker News Reader

Linux kernel RNG enhancements for 5.19

twitter.com

121 pointszx2c425 comments

Open articleView on HN

If MGLRU is merged, 5.19 will be a landmark as probably one of the most improved kernels in history.

What other notable improvements are there?

https://www.phoronix.com/scan.php?page=news_item&px=Linux-5....

https://www.phoronix.com/scan.php?page=news_item&px=Btrfs-Li...

https://www.phoronix.com/scan.php?page=news_item&px=Intel-IF...

But MGLRU alone is impactful enough to finally fix a multi-decade desktop limitation, so I don't even care about other changes.

For people who don’t know what MGLRU (Multi-Generation Least Recently Used) page cache algorithm is:

https://www.phoronix.com/scan.php?page=news_item&px=MGLRU-v1...

> There are wins reported for Cassandra, Hadooop, MySQL/MariaDB, Memcached, MongoDB, PostgreSQL, Redis, and improving the behavior in general for systems with limited RAM capacities or a lot of memory intense activities.

> MGLRU is already shipping with Google Android devices and also patched into Google's Chrome OS kernel.

That page didn't really explain MGLRU, at least to me. These pages however helped a lot:

https://lwn.net/Articles/856931/

https://lwn.net/Articles/894859/

An aspect I found interesting was that, in addition to the LRU aspect, it includes a PID controller to try to learn or compensate for its mistakes.

It is very exciting. Sounds like other filesystems under Linux now get zfs Adaptive Replacement Cache for free.

Will they bump the version number to 6?

Why would they? Linus changes version numbers on a whim, he stated that multiple times. Partly to make it obvious that all kernel releases are equal, partly for no reason at all.

The submitter is Jason Donenfeld, aka zx2c4, creator of Wireguard (and the `pass` password manager). This is an impressive record.

I don't like it when people touch the RNG and I think Wireguard glows.

Wireguard is probably the part of the kernel I trust _the most_ since it's < 5,000 lines of code and easily audited. I don't think the alphabet boys are involved in this one.

Have there been independent audits of it?

https://www.wireguard.com/formal-verification/

You are welcome.

Anything that involves cryptography has to remain suspect.

Yes, ultimately as there is no perfect and never break system. But pushing that is pointless as we need to have a safeguard now. What should it be? We can't wait for the God or Godet to come. We need a human solution.

And that might be faulty. Still, 5k lines are better than ...

It was working well before. And the Linux kernel has a history of highly pozzed, insecure RNG's that were given to the community by cryptographic "experts" as well.

It famously was not working well before, and no such "history" exists.

Reply on news.ycombinator.com