Illumos Dropping SPARC, Allows For Newer Compiler + Eventual Use Of Rust In The Kernel

[skeevy420]

I don't think it is the case that ZFS is the cause of this. That is my point.
I guess a simple way to test it would be to install libreoffice on a linux distro that uses ZFS.

Based on the numbers you posed, I think it'd be safe to assume that ZFS was the culprit. What you described and posted is how it is supposed to act. It just seems counter-intuitive if you come from the "I need every ounce of ram free when idle" mindset. I suppose running Ubuntu ZFS and doing the same things could test that theory out.

jacob While Ubuntu is pushing ZFS, no one else is adopting Zsys so almost everything Ubuntu does in regards to ZFS stays in-house. That sucks because Zsys is actually pretty nice and has a lot more potential than as just an Ubuntu thing. But more on topic, other distributions won't pick ZFS up because they're already invested in other tech. I doubt that the license is a factor in distributions picking ZFS or not. If that was the case then distributions wouldn't ship Nvidia drivers on install disks.

Red Hat has Stratis and other things that they're working on. SUSE is a firm BTRFS backer. So is Fedora. Debian doesn't back anything but it might as well be Ext4. Of the major players, all that's left is Ubuntu and they're using it. It just sucks that they're using it in an in-house manner that no one else seems to want to pick up...not even their alternates like Mint or KDE Neon (@ngraham *hint hint*) are using it.

By the time ZoL was ready for mainstream, most of the major distributions were invested in something else and Ubuntu wasn't. I think that's the simplest explanation of them all.

The problem with ZFS is there's a lot of demand for ZFS storage with minimal demand for ZFS as root. That's why almost every distribution has ZFS storage disk support and it's possible to install most of them to a ZFS root if one is determined enough. The demand is obviously there else it wouldn't be so easy to install everywhere. The problem is long term ZFS root maintenance. That can be a total motherfscker. Seems like only Ubuntu's Zsys is trying to tackle that on Linux in a mainstream, compatible for everyone, manner.

I feel like I'm talking in circles here so, yeah, that's my speculation on it all.

[pracedru]

Then limit ZFS's memory usage. That shouldn't be necessary with 4GB+ ram available. The FreeBSD recommended minimum is 4GB for comfortable use with most workloads and what you are experiencing is to be expected since ZFS should yield its used ram for the system.

I've read that in the past that 2GB is the ZoL extreme minimal a system should; FreeBSD says 1GB is extreme minimal (possibly with tuning); most places say 4GB+1GB per TB of ZFS storage is optimal.

I don't think it is the case that ZFS is the cause of this. That is my point.
I guess a simple way to test it would be to install libreoffice on a linux distro that uses ZFS.

[Volta]

I tried the new rfelease of Open Indianna Hipster and i was surprised at how much RAM it was using compared to an equivalent Linux desktop.
Does anyone know why OI Hipster is so RAM thirsty?

It was called slowlaris for a reason.

[jacob]

I wondered what happened to Fujitsu's SPARC64 processors, they sounded like real beasts like POWER 8 and maybe even POWER9 equivalents, of course AMD has turned the world on its head with 128 core systems in 2 socket systems but before that 4 way SMT was impressive to say the least.

I don't think it's a big loss if the open SPARC ecosystem didn't take off. SPARC processors like several others are famous for their sliding register window which sounds like a cool feature on paper but can turn out to be a real b!tch in real world code.

[kylew77]

Sparc was always the more interesting piece of the Sun pie, even if it was often poorly executed. Outside of ZFS, Solaris really didn't bring much to the table. They gradually added features that their competitors had for years and while it wasn't a "bad" OS it was almost always quite uninteresting.

Sparc on the other hand was an early adopter of 64 bit architectures (1995!), so early in fact they had to disable it on their first gen of products because of a cpu bug that would easily grant root. These days, everyone has 64 bit.

They took multicore/multtithread to a new level with their Niagara in 2005, with an 8 core 32 thread monster at 1.4ghz. Of course they shared some unfortunate components such as a single FPU for the entire chip, giving single core comparable performance to a 2 generation prior 450mhz UltraSparc IIe, but if you had an easy job like compiling, service static files over web or nfs, it was a monster. These days everyone does multicore+multithread but 32 threads on a single chip in 2005 was revolutionary.

UltraSparc IV would scale up to 72 sockets, up to 144 cores and half a terabyte or RAM, in 2005... Sure the previous generation could house more physical processors but they were single core, so this was a good jump. This scale of beast required a full 6 foot rack, giving you a level of cpu/memory capacity we can today fit easily into 1U... But in 2005 that was a massive system capable of running a single operating system rather than as a cluster.

Ultrasparc V and up were essentially Fujitsu products that Oracle happened to also sell, with notable improvements in 2012 and 2014, after which Fujitsu announced they were moving to ARM instead of doing Oracle's work for them.

I wondered what happened to Fujitsu's SPARC64 processors, they sounded like real beasts like POWER 8 and maybe even POWER9 equivalents, of course AMD has turned the world on its head with 128 core systems in 2 socket systems but before that 4 way SMT was impressive to say the least.

[CommunityMember]

Outside of ZFS, Solaris really didn't bring much to the table.

Dtrace? Solaris Zones?

[jacob]

Oh yes, going to a lot of effort necessarily guarantees that it must be worthwhile. Impeccable logic...

Linus' recent comments about ZFS were bang on the money—most of the "demand" for ZFS on Linux (or on any platform for that matter) is due to marketing and hype; not technical excellence. ZFS is a decidedly mediocre and inflexible filesystem.

Is there actually some significant "demand"? Ubuntu was trying to push it without much buy-in AFAICT and that's about it.

[jacob]

Technically speaking, Redox beat all of them to it.

Technically yes, but it's an experimental system. I was talking about mainstream established OSes: Windows, Linux and I guess Illumos qualifies too. The one still missing is Macos. The BSDs aren't going to happen IMHO for various reasons.

[skeevy420]

Another seat on the rust in the kernel bandwagon. I'll be interested in who will be the first to actually release it. My bet so far is on Windows, MS has been talking about it for a while now and seems to be the furthest down that road of all major OSes.

Technically speaking, Redox beat all of them to it.

[jacob]

Another seat on the rust in the kernel bandwagon. I'll be interested in who will be the first to actually release it. My bet so far is on Windows, MS has been talking about it for a while now and seems to be the furthest down that road of all major OSes.