Why APFS performance is bad on hard disks - it's the extents

[HDFan]

Larry Jordan has a great article about why you don't want to use APFS on a Hard Disk:

https://larryjordan.com/blog/apfs-i...campaign=12/04/17+Newsletter&utm_medium=email

He refers to this article which has nice visuals:

https://blog.macsales.com/43043-using-apfs-on-hdds-and-why-you-might-not-want-to

The bottom line - APFS creates a ton of extents, which on an SSD with fast access times isn't a problem. But a hard disk has millisecond delays positioning drive heads to access extents. This would result in (relatively) very slow disk access times if the disk were formatted with APFS.

 

[kagharaht]

I converted 2 portable external spinning drives (Seagate, WD) to APFS and 2 External Powered Desktop Drives. Refuses to mount with my Macs running HS unless you manually mount it with Disk Utility. I erased and converted the two portables back to HFS and they mount every time no issues. Thats one of the problems I've seen converting HD to APFS. I'll have to do the two desktop externals back to HFS to resolve the mounting issues.

 

[Fishrrman]

I have no plans to use APFS -at all- on ANY drives of mine for the near future.
HFS+ works for me.
Why change?
"If it ain't broke..."

 

[neliason]

Isn’t this only a problem for copied versions of a file? Also, isn’t this the same issue of fragmentation? Surely macOS can handle fragmentation issues for APFS.

 

[thejadedmonkey]

Isn’t this only a problem for copied versions of a file? Also, isn’t this the same issue of fragmentation? Surely macOS can handle fragmentation issues for APFS.

The file system uses something called "Copy on write", or COW. What this does is when you save a file, it only saves the changed parts, to another portion of the hard drive. So you have the original file, plus additive data that when combined, create the modified file. This way you can go back to previous versions of the file easily, and without it taking up much more space.

The problem is the sheer number of writes done to a hard drive will cause massive fragmentation. This is actually good for an SSD where multiple reads across multiple NAND chips makes things faster (all other things being equal). But on a spinning hard drive, the disk defragmenter won't be able to keep up: Right now, all I'm using my mac for is writing this response, and my I/O is 365kb per second. I don't know what system processes are doing that, but assume that each file write is also a file fragment... that's a lot of a system defragmenter to handle.

 

[decafjava]

Very interesting article, but from what I understood, those of us with hard or fusion drives are still using HFS+ even if we are running the latest version of HS. I think you can convert with a beta trial but I am in no hurry to do that and I hope Apple isn't either or I will be sticking to this version for a long time.

 

[neliason]

The file system uses something called "Copy on write", or COW. What this does is when you save a file, it only saves the changed parts, to another portion of the hard drive. So you have the original file, plus additive data that when combined, create the modified file. This way you can go back to previous versions of the file easily, and without it taking up much more space.

OK, I can see how this could be a problem. But could this also be a problem even With HFS+. Does every program on iMacs with HFS+ completely rewrite an updated file on save or is some sort of block comparison going on? For instance, if you edit metatag information on a large video (assuming the info is stored in the file data itself) does it rewrite the whole file?

The problem is the sheer number of writes done to a hard drive will cause massive fragmentation. This is actually good for an SSD where multiple reads across multiple NAND chips makes things faster (all other things being equal). But on a spinning hard drive, the disk defragmenter won't be able to keep up

What about Fusion drives? Being part SSD and HDD I would think they could handle this fine since the underlying Fusion logic tries to write to SSD first anyway. In thinking about Fusion isn’t it in the background constantly optimizing where data should be and moving it if necessary? If so couldn’t the same process fix problems with APFS on HDDs?