How did I .... ?

The 300Mbit/s ethernet is noticably a lot faster than the previous Pi but unfortunately only works for small amounts of data. This is a disppointment as I was hoping to run the Pi 24/7 to use as a native RISC OS SSD file server.

The RPI3B+ has an uneasy relationship with ShareFS, unlike the RPi3 which was good. (I think this issue was predicted by someone who understands this stuff.)

Titanium to RPi3B+ is good with ShareFSWindows of 1 on the Titanium and 2 on the RPi3B+. RPi3B+ to Titanium is beyond redemption. I do have a 100Mb/s USB to Ethernet adaptor that evades this issue, as soon as I can find it I might start using that again.

The problem is gigabit networking. The new pi has 1000Mb/s networking but can only transfer data from the network over the USB bus at 300Mb/s. The problem is when the remote device sends data at a sustained rate the pi device’s buffers fill up, packets get dropped, everything grinds to a halt waiting for retries.

ShareFSWindow 1 on the remote machine decreases the size and increases the number of chunks of data that are sent so gives the receiving device time to clear its buffers between bursts of data. Things can be improved by using flow control but not necessarily in the obvious way. The ArmX6 also doesn’t have full gigabit networking and isn’t USB but it does have a number of parameters to tweak relating to flow control which improved things for the ArmX6 a lot. Its early days for the pi driver so it’s likely to improve but it’s tricky to control the flow control using DeviceFS.

100Mb devices get around the problem by connecting to the other end of the wire at 100Mb and the switch you connect to takes care of the flow control to connected 1000Mb devices.

Colin,

The problem is gigabit networking.

I think I now understand your point. And, hence, why I am seeing no problems on my systems.

Although I do have a gigabit switch, I do not have any combination of gigabit systems that talk to each other without a 100Mb/s switch in the overall path.

There are basically four types of storage media:

You could add virtual media, as in cloud or what I call pseudo cloud. I regularly email encrypted copies of important files to an email address that doesn’t get downloaded, so should be there for years… though one ISP managed to delete all the old emails when it migrated its email system a few years ago. If you can stay within the sizes limits I wonder how long gmail emails will be retained?
n.b. I’m not sure that a paid for backup service is going to be any more reliable than using a free one!

I think it’s safe to say that there’s roughly zero chance that we’d all agree on the perfect layout of where to put stuff.

I think it’s safe to say that there’s roughly zero chance that I’d agree on the perfect layout of where to put stuff next week with where I think it should go today.

I should point out that optical storage is not something to be relied upon for long term storage unless the media are stored in appropriate conditions, which basically means cool and in the dark. There’s a fair amount of published practical research work to back this up. When the company I worked for used optical storage, we were aware of a tester that would create a picture of where the errors on an optical disc were located. The pictures were very interesting; you could see areas of the disc where errors were clustered together. Eventually we gave up optical storage in favour of hard drives in a RAID configuration, but remember that, in order to retain the high reliability (low failure rate), drives must be replaced when they reach the end of their rated lifetime, which is typically something like 3 years.

There is no truly reliable long term data storage system.