print slows down mid print

gloomyandy

@timtom said in print slows down mid print:

Scheduled moves 349773, completed 349773, hiccups 0, stepErrors 0, LaErrors 0, Underruns [6961, 13, 6362], CDDA state -1

The high number of underruns may be part of the problem, the question then though is what is causing the underruns? The read times from the SD card look to be fine, but it may be worth trying a different card if you have one, just to be sure.

TimTom

@gloomyandy

Thanks for the reply, still learning Duet troubleshooting.
Very weired, 10 MB

For everyone else running in this issue, @gloomyandy s report is documented here:
https://docs.duet3d.com/en/User_manual/RepRapFirmware/SD_card
Following documentation, I ran an SD card speed test in Duet:

Wow thats slow! It's the stock card, will replace these days (need to find such a small capacity card) . . .

jay_s_uk

@timtom why do you want a small SD card? I tend to use 16gb and 32gb cards

TimTom

@jay_s_uk

THAT IS SMALL!
I usually buy 128 or 256 GB in special offers, also mostly sandisk Ultra or extreme, so I can use them in Cameras, etc. as well. Small cards are comparably expensive (in price per GB).

jay_s_uk

@timtom ah, fair enough. FAT32 doesn't like large cards so not much bigger is supported

Phaedrux

@timtom said in print slows down mid print:

Free file entries: 8
SD card 0 detected, interface speed: 22.5MBytes/sec
SD card longest read time 4.5ms, write time 2.9ms, max retries 0

that looks about right for a Duet3.

Trying a new card is worth a try.

TimTom

@jay_s_uk
Sorry didn't want to be rude, I tend to pay a Euro more per card to avoid such kind of issues, currently the Sandisk 32 GB cards are only 2 Euros more than the "cheaper noname" cards. Filament cost and the time invest is worth much more than these 2 Euros so far.

@Phaedrux
Let's hope this solves the issue, cards should arrive today. Nevertheless, the card currently in use, is the stock card coming with the Duet board.

Will keep you updated.

TimTom

@phaedrux @dc42 @gloomyandy
I believe I found the Issue - small volume size lead to massive fragementation resulting in horrible SD-card performance.

In preparation for the "new cards" I took out the stock card and found out that the 8 GB card only has a 256 MB partition (so only 256 MB were usable).

Being curious I reformatted the card to it's full capacity and ran the Speed tests M122 P104 again, surprise, surprise speed tests dramatically improved:

with 256 MB partition (mostlikely massivly fragmented):

with 8GB partition:

Read speed nearly trippled!

Trying it again and BAM issue solved! - With the very same SD-Card!!!!

My assumptions and maybe what might cause this:

with 256 MB the "volume" is pretty full as my gcode+firmware files already occupied ~200 MB so not a lot of "free space"
doing a several firmware updates and lots of configuration changes, deleting and uploading always performs writes to the SD-card
FAT32 tends to fragment much more than other more modern filesystems
SD-card interface of Duet is not comparable with a high-tech PCIe SSD.

Bottom line the small volume got highly fragmented (explanation see below), resulting in super bad performance.
Unfortunatly I did not check the SD-card for fragmentation prior deleting, so I have no "proof".

Suggestion for Duet
I assume the small volume size of 256 GB is a result of historic artifact, because that allows to use any size of SD-card (down to 256 MB) in the production process. My humble suggestion would be

to change the manufacturing process to either start with lager volumes (e.g. 4 GB or 8 GB) or to make this a dynamic process, automatically sizing the partition/volume to the maximum available space (maybe add a cap at 32 GB to take FAT32 limit into account).
for existing customers and new customers receiving such cards, I'd put a clear Note about this topic in the documentation (something like "Consider reformating your SD-card to maximum capacity).

*Fragmentation explained by Roland (music instruments) -
When an SD card recorder stores new files, it squeezes them into any bits of space it can find. As a result, the data winds up scattered all over the hard drive. When the data from one song file, for example, is stored in many locations within the card memory, the card is said to be fragmented. When the recorder plays back a file from a fragmented card, it has to grab a little piece from here, another piece from there and so on. This is much harder than playing one continuous chunk of data from a single card location. This can slow things down unacceptably and lead to errors during playback, recording and/or any other function in the recorder.

Fragmentation is a naturally occurring process for a card used with any computer or recording device. SD cards that record audio are especially prone to fragmentation due to the start-and-stop nature of recording.

We recommend backing up all of your data and then initializing the SD card to clear out this normal fragmentation. To avoid receiving errors and/or problems, initialize the card a couple of times a year.. The following steps will guide you through the Card Backup, Card Initialize and Card Recover procedures:*

gloomyandy

@timtom The card supplied with the Duet3 has two partitions on the SD card. The first is a 256Mb FAT volume the second a Linux volume. The Linux volume contains the Duet RPI SBC image. This allows the SD card to be used either with a standalone system (using the FAT partition) or an SBC based system (with the SD card in the RPi). So the small size of the FAT partition is deliberate. As to if this is a good idea, I'll let others decide.

Phaedrux

After some investigation we've solved the issue of the too small partitions going forward. There were likely only a very small number of early Duet 3 boards that have the problem.

TimTom

@phaedrux
Well the board I had was not too old (I received it 1-2 months ago), it already had the current stable version on the SD-card.

As I'm a product manager myself (also for snom phones in the past = embedded devices), I'm very curious on what the solution is. Having invested some time and money (filament and new SD-cards), I'd love to hear what your conclusions are. Feel free to reach out on separate channel

Phaedrux

The board you got may have been produced much earlier than when you bought it.

The solution was just to change the amount of space allocated to the pi partition and the standalone partition so that the standalone isn't limited to such a small space.

TimTom

@phaedrux
Ahh ok, so you already adjusted the partition scheme, great to hear

According to serial my board was 2021-06.
Nevertheless I'd add a note to Setup/install guide, I'd written something but can't find my docu credentials at the moment.