Alternative USB driver/overlay for Raspberry Pis

[hamishmb]

Hi all,

As you may recall, we had some sporadic network issues in previous years. All seems fine this year, but I thought I would note that there is a new driver in development for the USB chip in the old (pre-v4) Pis, and this fixes many bugs. I have been using the driver at home for a few days now and have experienced much better stability. The only compromise is that it is a bit slower, but I haven't experienced this personally. Ethernet connects through USB (even for the Pis with an ethernet port integrated), so this also helps network reliability.

Noting this down here in case it's useful information for later. The link to the thread where I was discussing the driver is here: https://github.com/raspberrypi/linux/issues/3843

I was thinking that, pending some confirmation about how close to production this driver is, and me using it at home for a few weeks, it might be a good idea to pre-emptively enable this driver given our issues in the past. This is a simple change that is easy to apply and to revert. Thoughts?

[TerryJC]

Seems good to me.

[Penri]

Sounds promising, are there any before and after tests we could apply so measure the effectiveness of changing the driver?

[hamishmb]

Here are my thoughts:

1) We can check for kernel OOPSes and network warnings in the system logs of each pi to determine if we've had problems lately that have been automatically resolved.

2) We could transfer some large files between the pis before and after and measure the speed.

3) I have a spare SSD that I use to test a piece of commercial software I write (no data on it at all - it's a secure erase verification program), and a suitable USB caddy. We could plug this in to each pi of a different type that uses USB/is to be on the network (Zero, Model 1 B+ for VPN?, and aren't there some Pi 2s and 3s?) and see what the performance difference for each one is.

NB: Deploying this just on the river system pis (or even just a couple of them) at the beginning seems like a good idea - gives us time to find anything wrong.

NB 2: I think this is a relatively recent development. Other pis that are joining the network/depend heavily on USB may need to be updated to use the driver (or at least to use the version that I'm testing at home).

NB 3: This is not necessarily a priority task - we've (still) got some NAS debugging, code merging, and backup-solution-creating to do. Of course, we can multitask, but no need to put ourselves out updating things. Probably worth having a discussion about what pis we might want to use this driver.

NB 4: This all depends on what I hear back from the kernel guy on the thread, but they seem confident in the driver, so I reckon it will be positive.

[TerryJC]

We could plug this in to each pi of a different type that uses USB/is to be on the network (Zero, Model 1 B+ for VPN?, and aren't there some Pi 2s and 3s?) and see what the performance difference for each one is.

We only have Pi Zeros and Pi 3s on the network or in the WMT Site as a whole come to that.

Everything in the Railway Room uses a Zero, as does the Loos Sound Effects. As you know the River System SACs & Gate Valves all use Zeros . The Bells, Webserver and VPN Server all run on Pi 3s and the up coming Tiny-TVs will all have Zeros (although they will use WiFi not a USB/Ethernet Adaptor). The Visitor and Staff GUI will need at least a Pi 3 unless we strip down the desktop software and install a minimalist browser.

[hamishmb]

Okay, that makes things even simpler I guess.

[hamishmb]

Okay, seems like the only issues are performance in some situations. Once we test this for ourselves it should be fine. Issue is now marked as fixed on GitHub.

[hamishmb]

Unfortunately it just fell over, but with the same old problem. Not necessarily the USB drivers fault (it probably isn't) but still unfortunate. As I say, I'll test it for a bit longer and see what happens. My pi has other software issues at the moment so I'm not going to necessarily draw too many conclusions from this.

[hamishmb]

As for the testing with my SSD and caddy, all bets are now off because there has been a COVID spike in Wimborne and in Bournemouth - I will not be going in any more.

We can still do network-based testing though.

[hamishmb]

Okay, this still seems to be working really well for me. I could move a couple of pis over to this new driver if you're happy for me to do it. Which ones would I do first?

Also happy to do some network speed tests before and after.