Linux Tests Of The QNINE M.2 NVMe SSD Enclosure To USB-C Adapter

[wagaf]

Thanks for the great review !

[edwaleni]

On the ASUS PRIME Z390A board, the 2nd M.2 slot can accept a properly keyed SATA SSD, but it will consume a resource from the onboard SATA controller. You have to enable it in the BIOS.

The JMicron NVMe to USB 3.1 Gen 2 controller uses 2 PCIe lanes (v3) to communicate with the SSD.

Based on Michael's tests, the adapter is fully using the maximum bandwidth available via the USB port and controller.

What I am concerned about is the 870EVO performance in the M.2 slot alone.

The Z390 has a very unique approach to USB control, especially between Gen 2 mode (where it needs to detect voltage on the C port) and Gen 1 and lower (A ports).

If it detects voltage on the C port it uses 20:f0, if it doesn't detect voltage it uses 20:f1 via the xDCI interface.

It sounds like Intel has essentially placed a USB switch and voltage detection determines the route on the PCIe bus.

[torsionbar28]

Surely, you was not IT guy that time.... USB was a huge step forward over connecting the things over serial (or even parallel) port. Installing drivers for serial devices was troublesome and painful. Configuring programs to use them required much of configuration: setting baud-rate, number of stop bits, etc.

Sounds like you're confused here. RS232 devices do not require drivers. Their simplicity is why they are still in use today. Baud, data bits, parity, and stop are the only parameters. Not that you ever have to change them, as nearly everything uses 8/N/1. USB on the other hand, was a complicated cumbersome mess, where every device requires a driver, Windows users call this "driver hell". Remember when Bill Gates did a live demo of Windows 95? He plugged in a USB device and the whole OS crashed with BSOD, for the world to see. No such problems with RS232 serial. BTW my first PC was a Compaq 286 with 1200 baud modem...

USB was a huge step forward. USB 2.0 was finally something reliable and fast enough for most of the things of that time (maybe, not for you). Firewire was there but did not get widespread use. I never had a single device. Complain or not, it never got widespread use in circles around me, same as SCSI disks. Anyway, I am happy that you are finally happy with USB 3.1.

Incorrect, most pro audio/video equipment was firewire only. Because USB 2.0 was so crappy and slow, high bandwidth devices all came in Firewire versions, for people who care about performance. At one time, I had Firewire DVD burner, Firewire flat bed scanner, Firewire audio mixer, Firewire hard drives. It was a no brainer, really. If you like 30 MB/s, USB2 is ok for you. If you like 80 MB/s, Firewire delivers it. Did you think it was a coincidence that Apple used Firewire heavily, and also most media/design people used Apple? When you have large audio and video data to move around, USB2 was simply too slow.

But if 30 MB/s USB2 works for you, hey that's great for you. I'm just glad that after 23 years of waiting for v3.1, we finally have a USB that doesn't suck.

[PCJohn]

Sounds like you're confused here. RS232 devices do not require drivers.

I might be mistaken on this point, but how would you connect a printer? You need to install printer driver and setup port which the printer driver needs to use. But I might be mistaken here. I am happy that companies joined together and created standards, replacing whatever computer ports by some standard. Plug-and-play seemed a good idea to me even although it used to be called plug-and-pray. But for me it was right way to go. But I do not mind if FireWire would become widely used standard instead of USB.

Their simplicity is why they are still in use today. Baud, data bits, parity, and stop are the only parameters. Not that you ever have to change them, as nearly everything uses 8/N/1. USB on the other hand, was a complicated cumbersome mess, where every device requires a driver, Windows users call this "driver hell". Remember when Bill Gates did a live demo of Windows 95? He plugged in a USB device and the whole OS crashed with BSOD, for the world to see. No such problems with RS232 serial. BTW my first PC was a Compaq 286 with 1200 baud modem...

Feel free to like RS232, it is really simple from the signal point of view and if it works, it works. I still use one parallel port and it is just a pain. USB is a dream of heaven compared to my parallel port. My future is only any plug-and-play technology. A device should identify itself to the OS. This way, driver installation is much simplified. I would not like to go back. Driver-hell is probably not hell for me. I suffer by DLL-hell only. Surely, I accept that bad driver easily crashes OS, like it happened to Bill Gates. And no application or OS is bug free. But situation improved much. My first PC was 286 with two serial and one parallel port while I had hands on IBM-XT before.

Incorrect, most pro audio/video equipment was firewire only. Because USB 2.0 was so crappy and slow, high bandwidth devices all came in Firewire versions, for people who care about performance.

Oh, now I see where you came from. I was wondering why somebody might use so strong words against USB 2.0. Majority of us who are not audio/video would probably not call USB 2.0 crappy. Personally, I am also transfer-hungry (100GiB/s), but it is not towards external devices.

[torsionbar28]

I might be mistaken on this point, but how would you connect a printer?

We've strayed way off course now, if you review my original post, you'll see I was criticizing USB when used for storage devices. USB is perfectly suitable for low bandwidth devices like keyboards, mice, printers, etc. and I have no problems with it in this usage. This was after all, its intended purpose when it was first invented. But prior to USB 3.1, myself and many others have found USB to be entirely unsuitable for use with external bulk storage devices.[/QUOTE]

[starshipeleven]

USB 3.1 uses UASP for storage, so full SCSI command set is available. USB prior to 3.1 was junk, with junk protocols and implementations that were hit or miss on SMART. USB 3.1 fixes this mess finally.

I have quite a few sata drive boxes that support UASP but still don't show a damn thing on SMART on Linux (while everything works on Windows), so I'm unsure of what you are saying here.

[starshipeleven]

RS232 devices do not require drivers. Their simplicity is why they are still in use today. Baud, data bits, parity, and stop are the only parameters. Not that you ever have to change them, as nearly everything uses 8/N/1.

Fuck serial connections. Every single piece of shit industrial automation or high end businness device using a "serial" connection is NEVER EVER using a plain 232 but always some form of 232-alike stuff where they rewire the connector and place resistors between lines, and require random funky settings on speed and stuff, just because they can. Fuck them with a chainsaw.

USB on the other hand, was a complicated cumbersome mess, where every device requires a driver,

Yes. And this forces the manufacturer to actually deal with their shit on their own, so I don't have to set up shop and splice wires all the time. Any USB device has been painless on my side.

Windows users call this "driver hell".

Never heard the term, and I'm in IT since 2 decades at least.

Remember when Bill Gates did a live demo of Windows 95? He plugged in a USB device and the whole OS crashed with BSOD, for the world to see. No such problems with RS232 serial.

No such problem with punched cards or magnetic tape either, not a good argument for either.

Incorrect, most pro audio/video equipment was firewire only.

That's... not really widespread use in any way.

When you have large audio and video data to move around, USB2 was simply too slow.

maybe you don't know of its existence if you were in the Apple camp, but eSata was a thing for plebs and was common enough up and until USB 3.0 appeared. Also was Sata so SMART did work fine also on Linux 100% of the times.

Laptops had some EsataP ports https://en.wikipedia.org/wiki/ESATAp with combined esata and power for an external drive.

[torsionbar28]

That's... not really widespread use in any way.

McDonalds restaurants are widespread. Widespread != good. Some folks prefer 'good'. For everyone else, there's the dollar menu.

[starshipeleven]

McDonalds restaurants are widespread. Widespread != good. Some folks prefer 'good'. For everyone else, there's the dollar menu.

Don't shift the goalposts.
The point was "Firewire was not widespread", and "most pro audio/video equipment" is still a niche so even if Firewire did rule in this niche (I was there so I remember it did), you can't say it's widespread.

Also don't try to make Firewire look like a fine choice for fine people, it was used only because it was faster.
Most of its very cool features have always been irrelevant (I'm probably the only one left that actually knows what you could do with it, daisy-chaining, devices writing and reading independently on other devices in the daisy chain, sharing a network through devices in the chain, and others).
That's why it basically stopped being developed as soon as eSata appeared, and dropped entirely as soon as USB finally catched up.

[kcrudup]

(I understand this thread is nearly a year old, but this is the only such discussion on Phoronix, and hope this helps someone)

You don't really have to guess, since [the JMicron JMS583] is currently the only NVMe <-> USB 3.1gen2 chip on the market. German c't magazine recently tested two such external enclosures ... with similar results: The connection is simply unstable with all enclosures using this chip, and not only under Linux.

I had the same experience- I'd bought a few USB-C/NVMe adaptors from Amazon, and every one with a JMicron chipset would fail under heavy load- UASP is a multi-queue format, and any time I'd do anything that stressed the writes ("e2fsck", "compact partition" under VMware, etc.) the chipset would lose it and the Linux (5.5.-rcx) USB layer would have to reset the USB and dequeue all the queued requests. It would recover without going offline, but all I/O would pause for 5-20 seconds. It had to have been power related, as I didn't see this when it was plugged into the USB-C Gen2 port on my Thunderbolt dock (which has an 85W PS) and it worked well then, so I'll use it there exclusively when migrating NVMe drives)

I guess we'll have to wait for the ASMedia ASM2364 chip to hit the market before expecting different results.

One of the ones I'd tested, an adaptor from UGReen (https://www.amazon.com/gp/product/B07NPFV21K) has the ASM2364 chipset (174c:2362) and it's worked beautifully- no errors of any kind and excellent speeds (as Michael had shown in his testing, it's near-wire speeds). I've moved my VMWare .vmdks onto it now.

BTW, curiously enough there was a new controller type, the Realtek RTL9210 that I'd heard was supposed to be good, too- but it suffered from the same issues as the JMicron controller, and would get reset by the USB when the queue depth got too high. This, too must have been power related, as it also didn't do it when connected to my TB dock.

Thankfully Amazon has a generous return policy.