EXT4 "Fast Commits" Coming For Big Performance Boost In Ordered Mode

[AdamOne]

But which is faster for floppy's?

[linuxgeex]

But which is faster for floppy's?

mail:// is the best filesystem for those files. Use the eject button.

[andreano]

My takeaway from the slides: Making fsync work the way you thought / the way it should.

Before: A process fsyncing a single file had the side-effect of flushing every outstanding write on the same filesystem (including log spam, browser cache and whatever tmp files exist in this instant), and of course commanding the storage device to flush its write cache too, and penalizing this obedient process with the full latency for charter-ferrying all that baggage all the way down to the designated non-volatile storage.

After: The charter ferry is replaced with a private jet – we're not taking other processes' baggage with us to the storage device's write cache. But from there, it's still a charter ferry to the final non-volatile destination. Actually multiple if there are multiple queues, as with NVMe (if I understand the NVMe command set right).

[Azrael5]

Try something as simple as unzipping files, it's obvious.

The issues you refer deal with Microsoft operating systems succeeded to Windows XP. Compare Xp to any Linux operating systems in a mechanical hard disk so to verify what I state.In Xp the hard drive runs without any noise and it is quick, unlike any kind of Linux operating system, moreover, if the amount of ram is full, Linux operating system begins to be slow struggling a lot, not only, the operating system continues to access the hard drive also after the Ram is released. Here why developers are improving the I/O management and file system as well. However, make a test on your own so to be aware.

[Azrael5]

operations / using / in / Linux / to / swap / any / I / file / and / to / whole / back. / it / both / comparing / XP / the / the / run / when / overtakes / systems / made / slowdown / hard / simple / that / struggled / / dual / possibility / without / becomes / systems / same / been / RAM / efficient / operations / system / amount / drive / have / and / them / it / linux / the / used / how / I / in / comes / if / are / boot. / RAM / file / operating / linux / generally / operating / noted / the / not

I wanted to see if passing the above text through The Scramblinator would make it more clear, but no.

"I have noted that linux file systems are generally not so efficient comparing them to how XP makes the same operations simple, when I used to run both Xp and any kind of linux operating system in dual boot. Linux operating systems struggles in hard drive operations and if the whole RAM amount is full once the swap system is invoked, the slowdown becomes without any possibility the RAM comes back operative. A further issue I noted is that no noise is hearing during hard rive operations o Xp unlike any kind of Linux operating system"

check if the text is more understandable now.

[spirit]

Some operations are faster on Windows, like deleting a file or folder with files, I heard windows just deletes the file while Linux first deletes the nodes of a file and then the file itself. But lately file deletion speed on Linux seems normal, not sure.

with discard mount option, linux delete file data too, not only inode. And it's async since some years now. (previously linux discard was syncronous, so you were waiting for file data delete).

AFAIK, windows only do it through scheduled task, with optimize disk task. default delete only delete inode.

[oiaohm]

"I have noted that linux file systems are generally not so efficient comparing them to how XP makes the same operations simple, when I used to run both Xp and any kind of linux operating system in dual boot. Linux operating systems struggles in hard drive operations and if the whole RAM amount is full once the swap system is invoked, the slowdown becomes without any possibility the RAM comes back operative. A further issue I noted is that no noise is hearing during hard rive operations o Xp unlike any kind of Linux operating system"

check if the text is more understandable now.

Linux stalling out in swap is not a file system problem. Linux installed normally use swap partitions instead of swap files as well so truly not file system in the out of memory ice-station problem. Windows generally use swapfiles so this comes a different beast where disc io is going though file systems but this is not Linux

https://www.phoronix.com/scan.php?page=news_item&px=systemd-oomd-coming-soon

There is different work out there that can make Linux behave a lot better in once swap file is involved. Ram will come back with Linux once the OOM killer decided to get it back without something like oomd but most people are not willing to wait the 8 to 10 hours at times for the Linux kernel to decide to finally fix the problem.

[Azrael5]

Linux stalling out in swap is not a file system problem. Linux installed normally use swap partitions instead of swap files as well so truly not file system in the out of memory ice-station problem. Windows generally use swapfiles so this comes a different beast where disc io is going though file systems but this is not Linux

https://www.phoronix.com/scan.php?page=news_item&px=systemd-oomd-coming-soon

There is different work out there that can make Linux behave a lot better in once swap file is involved. Ram will come back with Linux once the OOM killer decided to get it back without something like oomd but most people are not willing to wait the 8 to 10 hours at times for the Linux kernel to decide to finally fix the problem.

In my case, the operating system continues to involve the hard drive instead of RAM although the RAM has been released when previously it was completely occupied. This behavior is not normal, how it is not normal that the hard drive looks like overloaded. However if the developers are improving the way the file systems run it means that they are able to make them better and more efficient than they actually are. If they were perfect since the beginning no improvement would be possible.

[oiaohm]

In my case, the operating system continues to involve the hard drive instead of RAM although the RAM has been released when previously it was completely occupied. This behavior is not normal, how it is not normal that the hard drive looks like overloaded. However if the developers are improving the way the file systems run it means that they are able to make them better and more efficient than they actually are. If they were perfect since the beginning no improvement would be possible.

I will try to make it more clear.

File systems and the system controlling swap/ram in Linux are two independent systems.

Using harddrive instead of ram is the swap system so improvements to ext4 will do nothing to the memory behaviour there. Linux does not auto pull everything back from swap once it been shoved in there. The behaviour of once something is placed in swap of the Linux kernel not instantly freeing it is absolutely normal and clearing items out of swap do take IO operations involving where ever the swap is .

To effect swap and file systems at the same time it would have to be alterations in the block layer subsystem that this ext4 alteration is not.

This ext4 is fast commit as in writing to disc after running software exceeding ram you would be after fast read from file system not fast write to repopulate file system buffers that would have been dropped due to lack of ram.

[mulenmar]

Some operations are faster on Windows, like deleting a file or folder

https://unix.stackexchange.com/questions/37329/efficiently-delete-large-directory-containing-thousands-of-files

Sometimes Perl is the sane solution. Yes, I know, crazy talk and all, but it's true.