anyone try two USB 3.1 ver 2 NVMe enclosures in RAID 1?

[whodiini]

I am considering a cheaper way to get high speeds on external enclsoures. One can get a decent NVMe enclosure for $20 that transfers 1 GB/sec via USB 3.1 ver 2 that is stable and reliable. So for $40, I can get 2 and in principle, putting them in parallel and running software raid 0, I should be able to get a throughput of 2 GB/sec for $40, This is 1/3 the cost of a thunderbolt 3 enclosure which has all sorts of compatibility issues and close to the same speed as well as being more reliable. So has anyone tried this? Only downside I see is that one needs 2 NVMe drives, preferably identical. I may try this out as an experiment,.

Comparisons:
SSK NVMe USB 3.1 ver 2 $18 1GB/sec xfer speed, very stable with latest firmware
OWC envoy express thunderbolt 3 $80 1.5GB/sec, very stable
2 SSK in raid 0 $36 2 GB./sec ???
Other thunderbolt 3/4 $120+ 2.5GB/sec, finicky with SSD, poor support

 

[Juicy Box]

putting them in parallel and running software raid 1

I think you mean RAID0, or striped RAID.

 

[whodiini]

I think you mean RAID0, or striped RAID.

yes, raid 0. Corrected

 

[Juicy Box]

I may try this out as an experiment,.

I have not done with with NVMe, nor over USB 3.2x2, but I have with SATA SSDs over USB2, USB3(5Gbps) and also over FW800.

I have also made SW RAID0 with mixed bus, such as one SSD over USB3, and the other on FW800 or one SSD on USB3 and the other on TB1.

I keep a lot of screenshots from tests, I will see if I can find them.

The following is what I found so far, everything is sequential read speeds, unless stated otherwise:

The two FW800 SSDs in RAID0 had a little improvement over a single SSD on FW800. 85MBps for the RAID0 versus 74MBps for the single SSD.

Two USB2 SSDs in SW RAID0 was only 20% faster than the single USB2 SSD. 48MBps for the RAID0 versus 40MBps for the single SSD.
Although, the Random read speeds more than doubled 15.2MBps for the RAID versus 7.6MBps for the single SSD. The Random writes where 50% faster for the RAID0m 12.3MBps for the RAID0, and 8.1MBps for the single.

The mix of the USB3 and TB1 doubled the single SSD speed. 762MBps for the mix RAID0, versus 386.5MBps for the slower of the two drives (TB1). I didn't test the random speeds, just sequential.

 

[whodiini]

I did some tests as an experiment. Here's a summary: for intel macs, it works great. For M1 macs, its more complicated and device dependent. There are issues on the Studio that may or may not be fixable via firmware updates.

Detailed answer:
For testing, I used the following 2 enclosures and drives I happen to have
1) SSk enclosure with JM583 controller 1.09 firmware. 500 GB Samsung 970 evo NVMe.
2) Netac enclosure with their controller and SSD.
Write read speeds tested with blackmagic. All drives formatted with disk utility as mac extended, journaled

Intel 2018 mac mini:
drives plugged into thunderbolt ports, using a single thunderbolt controller.
1) SSK 880 MB/sec write, 920 MB/sec read
2) Netac 840 MB/sec write, 910 MB/sec read
3) RAID 0 1500 write, 1560 read.

RAID 0 is 1.8x faster than individual drives! This is 2x cheaper than getting the cheapest thunderbolt 3 and performance is the same. $36 for SSK enclosures. It works.

Mac Studio with exact same drives and testing methodology
A. drives plugged onto the front USB ports on Studio
1) SSK 570MB/sec write, 600 MB/sec read.
2) Netac 700 MB/sec write 750 MB/sec read
3) Raid 0 720 MB/sec write 740 MB/sec read

WHOA! There are 2 issues here. The first is the major degradation in speeds compared to the intel mac mini. Something is wrong with the Studio M1 Max front ports. They are degraded. Issue 2 is that both ports seem to be run off a single USB controller so that the max speed is 10Gbit COMBINED from the front 2 ports! RAID is about the same speed as a single drive. This is awful performance off a $2k computer. I cannot believe that apple saved $2 not to have each USB port run at full USB 3.1 rev 2 speeds.

B. I then decided to run a test with the SSK disconnected from the front port and connected to my Caldigit USB 4 hub.
1) SSK into the caldigit hub: 620 MB/sec write and read
3) RAID 0: 1200 MB/sec write, 1200 MB/sec read

Much better, RAID is 2x the speed of a single drive.

But because of the degraded performance of the M1 Max Studio USB ports, 1200 MB/sec isnt that great. Would rather pay for the OWC envoy express $80 thunderbolt 3 and get 1500 MB/sec. Very dissappointed in the M1 Studio max USB ports.

 

[joevt]

B. I then decided to run a test with the SSK disconnected from the front port and connected to my Caldigit USB 4 hub.
1) SSK into the caldigit hub: 620 MB/sec write and read
3) RAID 0: 1200 MB/sec write, 1200 MB/sec read

What about 2 drives each to a different Thunderbolt port? I suspect ≈1500 MB/s.

What about 2 drives to a Thunderbolt 3 dock? or 2 drives connected to a Thunderbolt 4 dock that is connected to a Thunderbolt 3 dock? Putting a Thunderbolt 3 dock between an M series Mac and a Thunderbolt 4 dock disables USB tunnelling so that the inferior M series Mac USB controllers are not used. Instead, the Thunderbolt 4 dock's USB controller will get used. But a USB 4 dock uses a 10 Gbps hub so it's not good for RAID.

I think a Thunderbolt 3 dock like the HP Thunderbolt Dock G2 might be best. It's Thunderbolt 3 so it won't use USB tunnelling. Internally, the Thunderbolt controller of the dock has two USB ports - one for the downstream Thunderbolt port, and another for all the other USB ports which are connected using a USB hub.

Other Thunderbolt 3 docks with 10 Gbps ports might use an ASMedia ASM1142 which is limited to 8 Gbps instead of the full 10 Gbps.