A2 SD Card Speed Regessions/General Linux Question

Wronglebowski · July 22, 2019, 7:25pm

So I’ve been trying to educate myself more on Linux and I came across article linked on Hacker News.

ThomasKaiser/Knowledge/blob/master/articles/A1_and_A2_rated_SD_cards.md

# Performance impact of MMC storage used on SBC

## Historical background

SD cards were initially made for pure sequential workloads (storing/reading pictures or video). The vast majority of cards was only optimized for this single use case and performed extremely poorly with random IO access patterns.

When using SD cards as main storage on SBC (single board computers) this limitation resulted in poor performance since the main storage access pattern when running Linux or Android is random IO (accessing mostly small chunks of data in a random fashion in contrary to sequentially writing/reading large chunks of data in cameras and video recorders).

SD Association's 'speed classes' only dealt with sequential IO (minimum MB/s ratings) and it was an adventure to find SD cards that also perform ok-ish or even great with random IO access patterns as it's normal with Android/Linux. See here for a historical overview trying to find good SD cards for the use case 2 1/2 years ago: [https://forum.armbian.com/topic/954-sd-card-performance/](https://forum.armbian.com/topic/954-sd-card-performance/)

The stuff we do mostly on SBC requires high *random* IO but is not that much dependent on *sequential* performance unless you stream data (the NAS use case for example). So running off storage with superior random IO performance greatly improves the time needed to boot, browse the web (browser do constant random IO in the background), install updates and so on.

## 'Application Performance Class' to the rescue

Some time ago SD Association tried to address this problem and defined also specs and 'speed classes' for random IO access patterns mainly with Android in mind: the [Application Performance Classes](https://www.sdcard.org/developers/overview/application/index.html) A1 and A2 were invented.

![](https://www.sdcard.org/developers/overview/application/img/img_application.jpg)

### Application Performance Class 1 (A1)

This file has been truncated. show original

Basically, A2 SD Cards have a special caching mechanism that requires OS level support. Without it you see performance regressions back to slower than A1 cards. The Linux kernel currently doesn’t have this support, which seems to me like a big deal. My question is, how does this get resolved in Linux land? I understand anyone can submit code, but adding a driver like this doesn’t seem like a trivial bug fix.

FurryJackman · July 22, 2019, 11:47pm

The bigger problem right now is that SDXC is still using patent encumbered exFAT, and Microsoft left out exFAT for patents to be expunged for free use. That needs to be solved first before any SDXC card can work well with Linux without reformatting.