4-bit ECC in software is unacceptable

最新推荐文章于 2022-04-07 22:19:17 发布

转载最新推荐文章于 2022-04-07 22:19:17 发布 · 2.5k 阅读

文章标签：

#performance #flash #loops #each #windows #system

嵌入式开发常识专栏收录该内容

34 篇文章

订阅专栏

When ECC is calculated in hardware, performance is only minimally reduced.

In a test we did, write performance was reduced by only about 1.3%. By contrast, using software to calculate ECC in our flash file system using software reduced performance to a crawl:

Without ECC 600 KB/s 1-bit ECC 300 KB/s 4-bit ECC 5 KB/s (in software)

Clearly 4-bit ECC in software is unacceptable.

Even for a 512 MHz ARM11 we achieved only 30 KB/s.

There are several algorithms for calculating 4-bit (or more) ECC. BCH (Bose, Ray-Chaudhuri, Hocquenghem) is popular because of its improved efficiency over Reed-Solomon.

However, even BCH needs too many microprocessor cycles. A 256 KB flash block has 256*1024*8 = 2 Mbit. The ECC calculations (done for each 256 bytes) need 48 loops per bit, and for each bit it executes about 10 instructions. So totally it needs 2M*48*10 (about a billion) instructions to calculate ECC codes for one 256 KB flash block.

Even on a 2 GHz Windows PC, it needs about 400-500 milliseconds.