SDCard speeds with latest 5.29 ROM

Hi,

I updated my Pi4 from a 5.29 ROM build dated 6th Jan 2022 to the 30th Jan build and it seems that the sdcard read/write speeds have dropped.

Using HDSpeed on the earlier ROM I get as an example:

Average random variation from baseline = 8Mb
Average random variation from baseline 8387%
Block Load/Save, large file size = 8 Mb
Average block variation from baseline 5404%
Overall average variation from baseline 6995%

With the latest ROM:
Average random variation from baseline 7266%
Average block variation from baseline 5406%
Overall average variation from baseline 6398%
On average I get a 10% on average drop.

It seems the biggest hit is on Large file reads/1000 random bytes.

Any one else seeing similar changes between ROM’s?

How long ago was the first HDSpeed run? Can you restore the old ROM and check again now?

SD cards become slower as they age due to wear levelling, especially if the trim command isn’t used. When new any write is normally to a never used before block, but after a while all blocks have been used at least once, so it may have to do a copy and write to move frequently updated data to a less used block.

How long ago was the first HDSpeed run? Can you restore the old ROM and check again now?

So how about within in the hour then :-)

I tested the speed before changing the ROM from 6th Jan 2022 one and then after changing to 30th Jan 2022 and then reverted back all within the hour.

So I think I can see that the old ROM is about 10% quicker on average.

Um…

Um…

On the other hand I managed to prove that that the Pi ROM of 30Jan22 is faster that the ROM of 28Nov21 using HDspeed.

 Block Load/Save, large file size = 8 Mb
 Data transfer speed shown in Kb/Sec

                                ROM           28Nov21    30Jan22

   Save large block file                         5496%      5577%
   Load large block file                         5906%      6449%
   Save/load 50 Kb file                          3457%      5459%
   Save/load 512 Kb file                         3053%      6171%
   Save/load large block file                    5906%      6199%


  Average block variation from baseline          4814%      5710%

It was also apparent that HDspeed showed some variance of results on the same card, enough to call into the question the above.

Using other speed tests, including the time taken to unpack a ROOL source code tarball, showed no noticeable difference between the ROMs.

Hi David,

Thanks for testing.

I guess we can leave it here and it sort of does bring in to question the lack of a consistant performance tester for these sort of things but as it stands there are bigger things to fry as they say.

It isn’t just a RISC OS issue, testing filing systems on the Pi’s running Linux with fio often shows more than a 10% difference between consecutive runs. It’s only really useful to do when new and when the cards are getting old, as you can then see a 30% to 50% or larger drop in performance, especially with random access reads and writes, which indicates its time to replace the SD card.

There is a small chance this could relate to the SDIO changes that are coming in, as changes have been made due to the permissible combinations of SDIO controllers / channels on the various pi boards now that we can see/communicate with the other SDIO devices.

For example, Pi 0W and Pi 3B+ (ie. older design, but with wifi) has moved to a different SDIO controller for this reason. This is a slower controller for PIO operations (significantly) but performance was restored (and even surpassed in some cases) by enabling DMA on that controller.

As far as I know, the Pi4 setup hasn’t fundamentally changed – IIRC the Pi4 was the exception to the Pi SDIO weirdness – but if you do find repeatable (and I stress repeatable) performance regression, then please either raise a bug ticket here, or (perhaps better and) drop me an email.

For test purposes, the best way to regression test would be to test back/forth between old/new roms, with the same test suite. It can be difficult even then, as it is hard to say how much wear a given section of the SD card will have had. However, if repeat runnings and back/forth give repeatable/predictable results, then there may be some significance.

Please include version number of SDIOdriver module, and details of Pi model/version and ROM date.

I’m on the latest Pi nightly (5.29 6th Feb) on my Pi 4, and SDIODriver is still version 0.16 (19 Jun 2021). Doug’s message implies that the new SDIODriver is in the nightlies, but I’m not seeing it. Can anyone clarify?

Doug’s message implies that the new SDIODriver is in the nightlies

Really???

Can anyone clarify?

The new SDIODriver is not in the nightlies.

This story is entirely about SDIODriver 0.16.

HTH.

Yep, but since it was posted the same week as the annoucement of a new SDIODriver, it can be confusing :)

Yep, but since it was posted the same week as the annoucement of a new SDIODriver, it can be confusing :)

Indeed :) My bad.