[SOLVED] If you use a lot of fsync () at nand flash(TLC) rapidly decreasing the lifespan?

TeaYoung · 12-14-2015, 01:38 AM

Hello.

Do you understand English is not smooth, please ask.

SD Card(4GB) from my board fails, things are coming out frequently.

So is the SD Card (TLC) Debuging.

I've used the function of the [MMC debugging] in the Kernel.

Confirmed that the command was successful CMD13, CMD17, CMD18 command failed.

In circumstances where the issue is continuously recorded on an SD card, the current log.

The part that I suspect are part such as fsync ().

Following my hypothesis.

1. Even if the fsync () to write a 1byte is written is actually 4096byte because the characteristics of NAND Flash.

2. Repeat the process for a long time of 1 TLC, MLC as soon as this life is both touching.

Is this me this idea?

Below is the source code to test my hypothesis done. XD

Code:

#include <stdio.h>   
#include <string.h>  
#include <fcntl.h>   
#include <unistd.h> 

int main()
{
    int   fd;
    char *str = "1"; //1byte

    fd = open( "test.txt", O_WRONLY | O_CREAT, 0644);

    while(1){    

        write( fd, str, strlen( str));  //1byte write

        if ( -1 == fsync( fd))   //Write a 1byte (), but actually write the 4096 byte?  Due to the nature of the NAND Flash           
            printf( "fsync() failed");     
    }

    close( fd);
}

rtmistler · 12-16-2015, 07:33 AM

The more you write to a flash disk, the sooner it will wear out.

If you use that program which has no delay, infinite looping, it will eventually wear out a block on the flash disk, and then another, and another, and so forth.

What can happen is with write leveling, the different iterations of creating a new file can result in uniform writes to all free locations until you've largely written to all of the free locations one time, and then a repeat of that process.

So say you can write to any given location 10 times. And say further that there are Z locations. So it will either be:

Location A: Write first time
Location A: Write second time
... continue up to the 10th time
Now you can't write to Location A anymore, so you start writing to Location B
Eventually you can't write to Location B, so you move to C, D, ... Z
And now the disk is no longer write-able

Location A: Write
Location B: Write, and continue till you get to Location Z
Back to Location A: Write, and continue to Z a second pass
Eventually you'll loop 10 times and have written to all locations the allowable amount of times
End result the disk is no longer write-able

The more complex points here are:

Write cycle times are VERY HIGH in today's disk technologies
Given that they are so high, one write, ten writes, 100 writes, are not a big deal, it is partially a probability problem, when your limit is 2 Million writes, that's the advertised guarantee of some disks, and so some, maybe many locations may work with much higher write counts, and the manufacturer's intention is that very few, to no locations work with lower than 2 million write counts
Disks are very large, therefore whether or not they have block sizes for writes (they do) there's still very, very much data available for free space, provided your disk isn't largely full already.

I'm not sure writing one byte uses 4K, likely 256 bytes. Even so, 4K compared to 4G is a factor of one millionth.