RAID 0 with 3 drives only has 2x performances

@softraid-support 

Both are 0.7m, one is OWC, other one is Anker, both have TB logo + 3 at their ends.

This seems to be related either to the M1 processor or the 12.2.1 update:

M1 and macOS 12.2.1, one cable: 376MB/s (write), 1151MB/s (read)

M1 and macOS 12.2.1, two cables: 676MB/s (write), 1178MB/s (read)

i9 and macOS 12.2, one cable: 1013MB/s (write), 1147MB/s (read)

i9 and macOS 12.2, two cables: 1021MB/s (write), 1151MB/s (read)

@burningman 

It is true that the M1 is not as "storage fast" as intel processor computers, especially on USB.

It should not be that slow. I will do some checking, however.

@softraid-support 

Updated to macOS 12.2.1. Seems like M1 is the culprit.

i9 and macOS 12.2.1, one cable: 917MB/s (write), 1147MB/s (read)

i9 and macOS 12.2.1, two cables: 962MB/s (write), 1157MB/s (read)

I guess this is now something that is not related to SoftRAID, but to the OWC enclosure <-> how M1 handles TB3 connections?

@burningman 

Yes it is. But I thought the difference was much less. Looks like I have another round of basic performance testing to do.

@burningman 

Can you attach a SoftRAID tech support file:

When one cable is attached, and

When both cables are attached.

thanks

Could you change the title/move the thread to make it clear that the issue is related to macOS/M1?

I am currently transferring data so I won't be able to do many more experiments on the setup. However, I currently have ~900MB/s (read + write) (4 drives, RAID 4) while being only connected with one cable. The difference: The Thunderbay enclosure is now connected not directly but via a Thunderbolt 4 dock and spotlight is currently indexing the files on the RAID. Since I recall having the RAID connected to the Thunderbay at some point in my earlier test (where it didnt make it difference) I have the feeling that the additional spotlight index is at fault. Almost like macOS on M1 will throttle TB speeds depending on the number of processes accessing the device?

@burningman 

Its not easy to move threads, unfortunately.

Keep us informed, I have plans to do some cables testing and can test with a dock in addition to standalone.

Spotlight will kill write performance, until it is completed indexing. Perhaps that indeed was the cause of the slow results.

M1 is also not yet optimized.

@burningman 

I ran some tests and did not see the problems you described, at all. Performance was consistent on 3 and 4 drives, pretty similar on intel and M1.

M1 has a USB issue, but not much with Thunderbolt.

When I complete the tests, I may post them.

Posted by: @softraid-support

Spotlight will kill write performance, until it is completed indexing. Perhaps that indeed was the cause of the slow results.

Spotlight has finished indexing and numbers are roughly the same. Blackmagic and AJA give very different write speed readings though.

Which M1 model did you use? Did you have any other Thunderbolt devices attached? Which SSD enclosure did you use?

@burningman 

We find AJA to be more consistent.

I was using a Thunderbay, with 1TB OWC SSD's. it was direct attached. I used a M1 Air, iMac Pro as my main testing machines. they were direct connected.

@softraid-support 

So you didn't have any other TB components (display or a dock connected?)? Can you post the readings you got with AJA for 4x1TB OWC SSDs in RAID4?

Even on the i9 I only get ~2x read speeds instead of ~3x in a RAID4 config.

@burningman 

that is about what I found. I used an M1 Air, 1TB OWC SSD's (recent generation) and direct connected.

I don't have RAID 4 on the M1, I am enclosing R5, which is the same writes in all my testing, slIghtly slower reads than R4.

@softraid-support 

Ok, so your numbers are not so far off of mine. Assuming that the 1TB OWC SSDs also have ~500MB/s single drive performance, you are also only getting ~2x drive performance instead of ~3x in RAID4/5 (for 4 drives).

Isn't this indicative that something is wrong? Shouldn't you get read performance of around ~1.5GB/s? 

@burningman 

We need to re-optimize the drivers to get a boost. There are transactional costs where SSD's are faster than the I/O to calculate RAID parity. We are working on this.

However, yesterday, on the same computer I was getting 1K writes, I was seeing your problem of 350-400MB/s writes. And I tried "two cables" and yes, performance doubled (went to "normal:)

This looks like a bug somewhere, I need to research this and figure out why. it is M1 only, so it could be some interaction somewhere.