Approved: Fedora 33 Desktop Variants Defaulting To Btrfs File-System

[kloczek]

which is completely pointless since blocks are written randomly and blocks from the same file aren't necessarily contiguous

Sorry how did you get that?
ZFS is not writing blocks in random locations
Again you are guessing.

variable block size is only reducing but not solving the fragmentation.

Nope. It reduces amount of metadata about what is allocated. It reduces as well number of IOs
Again you are using intuition not hard knowledge.

Not sure how this is relevant with fragmentation

It is not at all because none of the ZFS users cares about fragmentation

]Ah, so it's like Windows 10 then, mostly closed source but at least gratis. Good to know.

Nope. Windows has nothing to do with ZFS.
Really ZFS is so different that whatever other experience you have it cannot be extrapolated on its area

[starshipeleven]

And because it is unbenchmarkable it must be slow

It sure was in the last benchmarks before that went up.

I don't know do you know that but after Sun was taken over by Oracle now only on Solaris kernel is working more people than during Sun time on entire Solaris distribution.
Every few months/SRUs is released batch of new features (I'm talking about only changes in kernel space).

Lies, people has been laid off and information about the situation also hit the streets

[starshipeleven]

And most of the file systems is even using that to maintain some read ahead data caches.

which is completely pointless since blocks are written randomly and blocks from the same file aren't necessarily contiguous

Sorry but ZFS does not preform what will be allocated using exact size of the block.
ZFS additionally has dynamic allocation using (record) which is from 1Kb up to 16MB.

variable block size is only reducing but not solving the fragmentation.

So you want to say that you don't know that COW transforms small random write IOs into sequential one?

Not sure how this is relevant with fragmentation

If you need Solaris you can use it for free.
Paid is support.

Ah, so it's like Windows 10 then, mostly closed source but at least gratis. Good to know.

[kloczek]

then why are you talking of OpenZFS as if it is so much different thant a core part can act so differently that it may have fragmentation problems?

Current sped of Solaris is unknown because it's Unbenchmarkable, aka part of the contract is that you cannot post benchmarks of it.
Keeping performance data secret does not give a good impression though.

Given that most of Solaris dev team has been disbanded years ago when it entered life support mode, I have my doubts.

And because it is unbenchmarkable it must be slow 😂😂😂
Try OpenSolaris you can dot with OS whatever you want
Oracle does not care about marketing of the Solaris and each ear it generates for only oracle more and more income .. end of the story.

I don't know do you know that but after Sun was taken over by Oracle now only on Solaris kernel is working more people than during Sun time on entire Solaris distribution.
Every few months/SRUs is released batch of new features (I'm talking about only changes in kernel space).

[starshipeleven]

(meta)SLAB alocatoor is core part of the ZFS from the beginning.

then why are you talking of OpenZFS as if it is so much different thant a core part can act so differently that it may have fragmentation problems?

Solaris was slow

Current sped of Solaris is unknown because it's Unbenchmarkable, aka part of the contract is that you cannot post benchmarks of it.
Keeping performance data secret does not give a good impression though.

If you will find today any symptom of the lower performance on Solaris compare to Linux you can open the bug report even without support contract and that report will be taken seriously.

Given that most of Solaris dev team has been disbanded years ago when it entered life support mode, I have my doubts.

[kloczek]

https://en.wikipedia.org/wiki/Slab_allocation

Slab allocation is a memory management mechanism intended for the efficient memory allocation of objects. Compared to earlier mechanisms, it reduces fragmentation caused by allocations and deallocations.

Key word REDUCES

Yep. Reduces below level below which defragmentation is necessary.

[kloczek]

lower (effective) IO? You know how fragmentation works?

The problem is that a "block", just like with an SSD memory cell or SMR zone, is a monolithic thing and I cannot just read parts of it, while a block is also big enough that it can contain parts from multiple files.

And most of the file systems is even using that to maintain some read ahead data caches.

If the data I need is split inside 400 blocks or if it is split inside 4000, or 40000 the performance changes because now I need to read more blocks to get the same data.

Sorry but ZFS does not preform what will be allocated using exact size of the block.
ZFS additionally has dynamic allocation using (record) which is from 1Kb up to 16MB.
This is why SLAB allocator needs to be involved.
Look on slabtop command output on Linux or /proc/slabinfo content and you will be able to find that within each SLABs size of the allocation using is fixed and different.
Looks like you still don't know what SLAB allocatior is.

Not saying this is a major issue like it is for trash-grade filesystems like NTFS, just saying that having something to deal with it if it happens, that does not involve copying terbytes of data to an external drive would be nice. Because it still can happen, for some workloads.

Can you lend me a Slowlaris license? Because if the only way to test your claims is to pay Oracle for a businness license I'm not doing that, I'm too stingy.

So you want to say that you don't know that COW transforms small random write IOs into sequential one?
None of the current filesystems has 1:1 relation to read IOs on VFS and block layer.
I have no idea how it is is with NTFS and NTFS is not subject here.
If you need Solaris you can use it for free.
Paid is support.
It is like this with Solaris and all OpenSolaris derivatives.
To download and use regular Solaris you don't need to pay anything.

[kloczek]

None is using Slowlaris, when everyone talks of ZFS they really mean OpenZFS now, old man.

I find it hard to believe that a core filesystem feature like that was changed only in the relatively recent past, in the proprietary ZFS fork that none uses because it's on Slowlaris.

Sorry you are talking about some fantasies.
(meta)SLAB alocatoor is core part of the ZFS from the beginning.
Look on "zdb -vvv" output and you will be able to find SLAB alocator metrics data.

Solaris was slow compare to Linux before Solaris 10 in which any possible to observe lower performance on Solaris was treated as critical bug.
Solaris 10 GA has been released 19 years ago.
If you will find today any symptom of the lower performance on Solaris compare to Linux you can open the bug report even without support contract and that report will be taken seriously.
Problem is that today Linux is quite often slower than Solaris especially on HW with high count of the CPU cores.
You would know that if you would be using Solaris at lease one time in your life.

[S.Pam]

Another great FS with no recovery tools whatsoever. Once you have a serious enough crash, or you delete files accidentally, say goodbye to your data. ext4 might not be modern, or fancy or may lack cool features but it's extremely reliable (every Android phone uses it - two billion devices).

There are plenty of tools to recover. But undeleting is not often possible. - But hey, this is why you use snapshots!

[starshipeleven]

Just please at least try to google for "zfs slab allocatior" then read first few links.

https://en.wikipedia.org/wiki/Slab_allocation

Slab allocation is a memory management mechanism intended for the efficient memory allocation of objects. Compared to earlier mechanisms, it reduces fragmentation caused by allocations and deallocations.

Key word REDUCES