Hacker News new | ask | show | jobs
by NavinF 1466 days ago
Oh wow I thought you were joking, but they’re literally 3 generations behind. I was recently complaining about the slow adoption of Wi-Fi 6E in phones, but now I know that true despair looks like.
1 comments

True but Linux and Mac is also one generation behind, but in a much more important field...filesystems ;)
is that so. freebsd and linux share the same zfs code and linux also has btrfs.

like... where is linux behind?

Soft updates!

It's a dead-end design, you say? Bah, that's quitter talk!

>It's a dead-end design

That's like saing journale'd filesystem are dead-end design, you don't make any sense.

No sure if you don't know it, but Netflix uses UFS2 and FreeBSD as storage on their CDN[1], on the other hand, for the easy and unreliable stuff aka containers Linux is acceptable, just restart it. ;)

[1] https://openconnect.netflix.com/en/appliances/#software

https://www.usenix.org/legacy/publications/library/proceedin...

>>Journaling Versus Soft Updates: Asynchronous Meta-data Protection in File Systems

>>10 Conclusions

>>Soft Updates exhibits some side-effects that improve performance, in some cases significantly. Its ability to delay deletes is evidenced most clearly in the microbenchmark results. For the massive data set of the Netnews benchmark, we see that Soft Updates' ordering constraints prevent it from achieving performance comparable to the asynchronous journaling systems, while for the small Postmark dataset, Soft Updates backgrounding of deletes provides superior performance. The race between increasing memory sizes and increasing data sets will determine which of these effects is most significant.

It's not that it isn't a fantastic design—I believe as much as the next person that UFS2 is basically best-in-class as a "traditional" dumb filesystem, and that soft updates is fundamentally a Right Approach. I wouldn't have mentioned it if I didn't!

But it's hard not to see that it's darn near barren ground for further work in filesystem design. And it's easy to see why if you scan the first paper—just look at the dependency flowcharts! Fundamentally the design is ... hard for mere mortals to work on. It's practically a miracle that it even happened once; nobody is keen on starting any new implementations.

I choose to fault filesystem implementers for not being (more) superhuman for this failing, obviously.

I’m not sure where the legend about softupdates being more complicated then journaling came from, but even comparing number of lines of code between SU and XFS journaling is enough to disprove it.
>btrfs

Haha, have fun with that Clown College of a Filesystem.

ZFS is not part of Linux.

The open-source IllumOS code in FreeBSD had fallen behind Linux ZoL (ZFS-On-Linux). FreeBSD adopted ZoL to keep up with new bugfixes, enhancements, etc.

https://www.freebsdnews.com/2019/01/10/zfs-on-freebsd-zof-is...

So, for open-source, ZoL is the 'mainstream' ZFS.

>So, for open-source, ZoL is the 'mainstream' ZFS.

Wrong, there is no ZoL anymore "just" OpenZFS:

https://openzfs.org/wiki/Main_Page

>>The OpenZFS project brings together developers from the Linux, FreeBSD, illumos, MacOS, and Windows platforms. OpenZFS is supported by a wide range of companies.

And again ZFS is NOT part of Linux.

I run ZFS on my linux servers and my linux NAS. It's served me well.
ZFS is not part of Linux.
Who cares? Nothing is part of Linux. It’s part of Ubuntu.
>Nothing is part of Linux

Like ~every other Filesystem? Not sure if you just lack that knowledge.