The only issue could be the speed, under 400MBps, while Finder is able to achieve 800MBps (in our case, reading from 4x HDDs Raid 5 and writing to a NAS).
RUBEN SALAZAR, 11 Sep 2022, 20:59
And before you got 700MB/sec with FFS according to your screenshots?
@RUBEN SALAZAR:
BTW: when you make performance measurements/comparsions, *restart* your mac between each run. Otherwise the numbers will be garbage due to buffers at various levels (OS, and hardware).
I have been testing, the app is stable in M1. Finder is still working faster, but I guess that verification between files 'is not free'. So close to 400MBps could be great for one touch process!
FFS 11.26 BETA 3 APP VS FINDER 220912.jpg (283.83 KiB) Viewed 4281 times
In the picture, Data sent with Finder, stopping, and syncing the same content with FFS 11.26 b3
And, yes, restarting computer before testing! THANKS! Any other improvements are welcome, and I am available for testing them!
Interestingly, I'm not able to reproduce the perf-degradation when OS buffering is disabled. Quite contrary, FFS is significantly faster (about 30% shorter runtime) for local SSD-based file copying.
But there's a different bottleneck. FFS could get rid of needless intermediary buffers. This gives another good speed improvement in my tests. I believe we can match Finder with the following version which even has OS file cache disabled for both source and target.
Thank you for this 5th beta. The behaviour is similar to 3rd. And winner is 4th!
FFS BETA 5 vs BETA 4 220913.png (451.96 KiB) Viewed 4224 times
In the picture, first installed Beta 5 and test, 400MBps average. Then overwritten with old Beta 4 and run, 600MBps average, the same as Finder copying-pasting
After testing 5th, re-installing 4th and testing, I re-installed 5th again and same result, going down to 400MBps again:
FFS BETA 5 then 4 then 5 220913.png (452.25 KiB) Viewed 4224 times
In the picture FFS Beta 5 second round. After first round I overwrote with Beta 4 getting the best performance.
Hm, this clearly needs more testing. I don't have other ideas at this point, other than maybe if the terminal command "cp" is as fast as Finder, we could analyze what clever thing it does: https://freefilesync.org/faq.php#trace
File copy while skipping the OS file cache is the default on macOS (both Finder and copyfile() function), so this seems to be the right direction. Disabling cache only for one side will problably not help in scenarios where source and target devices are reversed, so it's all or nothing. Therfore I'll include beta5 for the next release. If feedback is such that performance suffers in most cases, we can still go back to OS caching + system instability.
After further testing it became clear that buffer size might be the remaining issue. My SMB network share advertizes an f_iosize of 7405568 bytes, but for optimal speed matching Finder this needs to be 32 MB at least! I have no idea why buffer sizes that large make a difference, but apparently each network write is unusually expensive.
I've doubled the f_iosize sizes, which gets me in the range of only 5% slower than Finder in my tests.
With 4 x f_iosize I can match Finder. But I don't want to deviate too much from what is defined as "optimal transfer block size for the file system". Not so optimal, aparently. What do you get?
The behaviour of slow / fast is proportional to the use of the main storage, or the use of a SSD for testing.
The main is 4x HDDs in Raid 5 (thunderbolt 3), to a NAS (1PB) over 10Gbps network with Mac Minis M1, with 10Gbe adapter.
Pending the test with 8-bay Raid 5 disks.
Protocol is SMB 3. We have been forcing SMB 2 in the past (till 3 days ago) to avoid crashes with M1s' data transferring to other M1s.
But NAS is using 3.1.1.
Using Big Sur and also testing Monterey and Ventura. Speed behaviour is the same with beta versions in the 3 O.S.
Only happens with Minis M1. iMacs work fine!
Also pending the tests with Jumbo MTU. Now is 1500. But have to change several switches / ports
I just discovered this thread, so glad to hear this is being worked on.
I am having this issue on a Macbook Pro 16, Max 32core, 64gb ram.
My storage is a Synology 1621+ NAS over 1gbe for now.
I can help with testing from now on if needed.
I just discovered this thread, so glad to hear this is being worked on.
I am having this issue on a Macbook Pro 16, Max 32core, 64gb ram.
My storage is a Synology 1621+ NAS over 1gbe for now.
I can help with testing from now on if needed.
antag, 20 Sep 2022, 08:35
Perfect! Can you test betas 6 and 7 and compare the speed with Finder? Be sure to "sudo purge" before each test!
A curiosity... A Mac Mini M1 running 1 instance of FFS is getting about 250 or 300MBps from an external SSD connected through USB 10Gbps to a Thunderbolt 3 4x Raid 5.
And a second instance running at the same time, from other SSD to the same Raid, different folders, is achieving close to the same speed.
The sum of 2 FFSs running at once is faster, getting the speed of the bus.
We tested Beta 7. Speed is mostly at 450MB witht up to 750MB (with the finder/testtool we reach 950MB via smb). We just copied 271GB to the server without crash.
Occasionally ffs is hanging with 0MB for around 3-5 seconds and then ffs goes on with good speed.
So it is much better because we have no crashes and faster speed.
Our System is an 10GB Ethernet M1 Studio Ultra with 64GB Ram copying to a synology server via SMB with write speed up to 950MB.
So big thanks for the test and betas.
Is there anything we can help to test? Then we need a little manual or to do-list, what we should test.
But thanks for this great tool and the betatesters here. You save us a lot of time!!
Chris
I have tried all the betas and the latest 11.26. They all crash my Mac Studio when writing to my QNAP NAS. It is set up in RAID 0 for speed. File transfers quickly get slower and slower, from over 400 Mb/s to less than 100, and eventually the Mac hangs.
