What is the best file system for sharing files?

Question

linux

Please tell me which file system is better to use on a linux server that distributes static files, but also contains mysql databases and dynamic content?
I've heard a lot of nice things about XFS, but at the same time a lot of people are praising ext4 as well.

Answer 1

First of all, never use reiserfs as its development and native support in Linux is frozen.
In general, I use ext4 and xfs in production, they work about the same with the noatime options, there are both large and small files.
The one who writes that xfs is only for large files has not looked at her changelog for a long time. Why would ext4 be average with large files? It works great with large files compared to ext3.
If you have a rhel based distribution, feel free to use xfs there native support for it, if debian based use ext4.
Do not believe that the controller and disks decide, some caching in nginx + noatime options will give the cheapest sata disks more than a $1000 controller + sas disks without them.

Answer 2

XFS is suitable for large files of several GB, ReiserFS is suitable for a bunch of small files. ext4 is something in between.

Answer 3

If there are many small files, then ReiserFS.
Or, for example, we have a lot of small media files (music, photos) and a lot of clients, then fragmentation, not FS, will be a significant factor for us.
But here you also need to remember that in Linux (and not only) files are cached, and if there are enough RAM, then which file system will no longer matter.
In general, I think that in each case it is necessary to select.
I also think that if you are not an employee of the Google Data Center, then you should not bother with the file system. For the increase is unlikely to be more than 20%. But it’s worth it to get confused on the physical characteristics of the HDD, raid and RAM.

Answer 4

ext4 is a smart choice for today. Don't forget to correctly set stride, stripe-width if over raid, except for raid1.
To do this, for example, you can use this useful thing: busybox.net/ ~aldot/mkfs_stride.html
Also, the size of a chunk in a raid, depending on the data, often makes sense to set it different from the standard one.

Answer 5

Thanks everyone for the replies!
I'm leaning towards ext4 + noatime.

Answer 6

It was correctly written above: hardware solutions give the most tangible increase. That is, if you have fast disks, a controller (this is an option, because there are other options), the type of FS will mean relatively little.

Answer 7

In the general course of the question, I want to ask if anyone had experience comparing the performance of UFS vs. ZFS?