Performance of encrypted APFS RAID5 fixed?

No, it is not addressed.

This is a more complex issue, likely requiring macOS driver changes also.
Its a top priority now for next major SoftRAID update, if we can get this fixed, we will.

Posted by: @softraid-support

↑

No, it is not addressed.

This is a more complex issue, likely requiring macOS driver changes also.
Its a top priority now for next major SoftRAID update, if we can get this fixed, we will.

 

Does this help?

diskutil apfs defragment volumeDevice enable

@john-graham 

I doubt it , but I can test this. I generally test with cleanly created volumes, so fragmentation should not be an issue

I assume this also applies to APFS/RAID issue with SSDs as described here? (Write speed 380mb/s, read 512mb/s):

https://forums.softraid.com/creating-volumes/error-128-when-creating-raid4-on-softraid-pro-7-6-cfbundleexecutable-property-executable-cant-be-found/#post-20056

@burningman 

Here are some RAID encryption performance tests. These are on M1, with a Thunderblade 4.

I listed both 16k Stripe unit sizes and 64k, as there are significant differences with encryption

                        APFS        APFS encrypted (Mb/s)
                  Write   Read             Write   Read
RAID 0        2847   2969             2466   2182
RAID 4        1777   2651             1065     896
RAID 4 (64) 1910   2653               302   2092
RAID 5        1124   2606               933     912
RAID 5 (64) 2062   2520               285   2162
RAID 1+0    1406   2967             1028   2020

In essence, if you are using/requiring encrypted volumes, RAID 0 is your best choice, with frequent, scheduled backups. RAID 4/5 are OK performance, but if you need 1500 MB/s, then you want RAID 0

I also tested SoftRAID RAID 0 vs Apple RAID 0 and they are similar performance (Apple RAID does not support TRIM, so you will get much faster wear on your devices)

SSD's are slower than NVMe, I did not have time to perform standard SSD testing.

@softraid-support Thanks for hint regarding the stripe size. Are you numbers similar to SoftRAID 7 or have they improved by upgrading to SoftRAID8?

@burningman 

The performance will be the same, except for PCI cards (Accelsior 4M2), which will have better performance.

@softraid-support Thanks. Regarding the stripe size: I currently have 64K (RAID4, SSD) and now I'm tempted to switch to 16K. What is the best way to check how performance would be affected if I switched it to 16K?

I assume converting stripe size is not an option without destroying the whole RAID? Can I test it out by temporarily creating a new volume on the parity drive and have it rebuilt later?

@burningman

On flash media, you are unlikely to see any difference. The only way you can do a test, would be resize the volume smaller by 1TB or so, create a new 16K volume, test, then delete the volume, and later resize (expand) back to the original capacity. I don't think it is worth it.

@softraid-support Were you able to test this?

I'm very curious about this now, because I'm running out of options.

I'm trying to create an encrypted volume for a ThunderBay that I want in RAID 5, with 48 TB of usable capacity.

I've configured the ThunderBay as an HFS Plus RAID 5 within SoftRAID, but on macOS Sonoma 14.4.1, I've found that:

1. VeraCrypt doesn't really work--it freezes after a couple of days trying to create a 48 TB encrypted volume.
2. macOS Sonoma doesn't appear to be able to create an HFS Plus encrypted volume within Disk Image [the GUI] or `diskutil` [the CLI binary]. Sounds like Apple has eliminated the option, because they want to nudge users toward APFS.
3. If I just create an encrypted HFS Plus `.dmg` via Disk Utility, the performance is atrocious. The mounted disk image only seems to be able to write data at 5-10 MB/s.

So I'm curious about just using APFS anyway. I read this Knowledge Base article, and I watched the relevant section of the linked video, but I'm wondering if perhaps it's outdated because it's from 2018?

I was poking around with `diskutil`'s `man` page and and saw another user trying to enable defragmentation for an HDD, and saw another user on the Apple forums who was attempting the same thing.

It appears that defragmentation for APFS volumes is disabled by default, at least for the SSD boot drive, which makes sense so as to not waste CPU cycles on SSDs, but one would think that it would be enabled by default for HDDs, or at least Apple could provide some documentation for advanced users that defragmenting is intended for HDDs.

Could you perhaps consult with Tim Standing and get his Python script he describes at 8:00 to fragment a file and then test the automatic defragmentation via `diskutil`? If you don't deliberately fragment the files, and if you're only testing on an empty volume, you won't really be able to see any performance issues at all.

For what it's worth, when he was playing with the `defragmentation` option for `diskutil` at 12:55, he didn't notice any difference, but that was back in 2018. So perhaps things have changed.

@seth-goldin 

Nothing has changed. Defragmentation command still does not do anything. Neither does it address the fundamental issue with APFS of lacking "Copy on write"

On the good side, we are working on a way to get encryption on an HFS volume. We are testing and working on instructions. It will requite 10.15 to set up, but will work afterwards (i.e, a feature was removed in Disk Util)

When we have something I will post it and we will write up an FAQ

@softraid-support This does not sound like a great workaround. We're phasing out the use of Intel machines, since they're all EOL, and emulating Catalina on Apple silicon is difficult, if not impossible.

@seth-goldin 

Its only to set it up. Apparently Apple removed this "feature" in Big Sur, after being setup, it will work fine.

@softraid-support I understand that you're saying configuring encrypted HFS Plus is possible with Catalina, but it's not ideal because:

1. It requires old hardware, all so old that the units are beyond warranty from Apple;
2. The vintage products that can be serviced will continue to slip into obsolescence.

There seems to be conflicting anecdotes out on the Internet that enabling the automatic defragmenting does actually help for HDDs on APFS. Back in 2019, this may not have done anything, but at least one user claims that in August of 2023, it did have an effect, so it's possible that Apple has addressed this.

Can you confirm that you're seeing no effect within Sonoma? You mentioned earlier in the thread that you're testing on empty volumes, which doesn't really prove anything one way or the other.