RESEARCH REPORT: How dangerous are media errors?

I’ve been worrying about media errors: what if I store my CD collection on a DVD+R, lose my originals, then have a poor-quality backup?

So I did some tests.

MEDIA ERRORS vs. HIGH-LEVEL ERROR-CORRECTION

First, I wondered about the PI/PO errors that programs like KProbe report. I burned four DVD+R (TDK 4x) with my LiteOn 811s. Afterwards, KProbe reported some 1500 PO errors and some 500000 PI errors. This is considered a good result, according to the LiteOn forum.

However, remember that the file system and/or operating system (e.g. Windows) uses error correction to detect and fix occasional bad bits. After burning some 4GB onto each of my four discs, Nero verified the burn and claimed each to be successful.

Being suspicious, I copied the files from my DVD disc back to the computer. I wrote a short Java program to compare the files, bit by bit, and indeed: there were no errors whatsoever!

SENSITIVITY OF COMPRESSED FORMATS: MP3 and DIVX

Next, I worried: what if I occasionally do get a few bad bits, even after error correction?

So I wrote another program to randomly insert an error in an audio or video file every five seconds. I tried two types of errors:
BIT FLIP: A single bit was change from 0 to 1 or from 1 to 0.
BAD SEGMENT: I set one or more consecutive bytes to zero.

For an MP3 audio file, I could hear about one in ten bit flips. When I heard them, they were brief squawks or noise bursts (about 1/3 second long). These are about as annoying as scratches on old vinyl records: I can live with a few. Say if I hear one per song; then, since I can store some 1500 songs per DVD disc, I can live with some 15,000 uncorrected bit flips per DVD disc.

When I zeroed bytes, the sound effect was about the same. For a single byte, I still heard only 1 in 10, but when each bad segment was 128-1024 bytes long, I could hear every one, though the glitch was a little more gentle: just a brief absence of sound, which is less annoying than squawks.

For DivX video files, I noticed the errors resulted in three types of effects:

Type 1: A very brief flash or a very brief coarse graining (big, square pixels).
Type 2: Video freezes (0.5-2 seconds), while the audio keeps going, then continues just fine.
Type 3: WMP jumps back to the beginning of the file and stops.

For single bit flips, I noticed one in three errors; half were Type 1 and half were Type 2.
For 1-10 zeroed bytes, I noticed about one in three, mostly of Type 2.
For 128-1024 zeroed bytes, I got Type 3 errors within one or two minutes.

For kicks, I tried PDF document files. Single bit flips were bad enough to completely damage an entire page about half of the time. By damage I mean that the page was either blank for garbled.

SPECIFICATIONS

The audio files I tested were compressed with CDex, using MP3 at 128 KB/s rate.
The video files were encoded with FairUse, using 2-pass DivX + 128 KB/s audio and about 6 MB/min.
Both audio and video files were played back on Windows Media Player (WMP).
Computer: P3-933 with Windows XP and WMP9.
Media: TDK 4x DVD+R discs (Ricohjpnr01, $27 for 15 at Costco).
Burner: LiteOn 811s, bought new 1 month ago (February 2004).

CONCLUSIONS

I’m not too worried about the integrity of my files. I worry somewhat more about DivX movies, as a bad segment may cause WMP to stop playing the file; however, this may be a quirk of WMP: other players may be smart enough to skip a few seconds and continue.

Note that MP3 and DivX are aggressively compressed with little redundancy. If you store in less compressed formats, or even in a raw format, you are unlikely to notice any errors: who would ever notice a single bad pixel or a single bad audio sample?

When I run out of this first, pricey, batch of blanks, I plan to buy the cheaper Ritek G04. I’m curious to see how they hold up.

  • Jon Absoul

PI/PO errors are not file system errors, they are completely different. The KProbe measurements of the errors are before correction is applied in the drive, so there are no errors being seen by the OS.

The DVD specs are fairly clear on what level of errors that a drive is expected to correct, but this can vary greatly from one drive to the next, and from one reading to the next. Given the fairly extensive history with CDR’s and read errors, it’s logical to assume that those discs with high error rates are far more likely to fail in some way as they age. But only time will tell this. Unfortunately, we didn’t really know about the problem with CDRs untill they started failing, and the same will be true for DVD.

The only time you will see data errors on a disc is when the PI/PO rates exceed the drive’s ability to correct them, which only occurs on a really crappy disc. So, using data errors, or the lack of them, to measure a disc’s “quality” is foolhardy. This is why we have tools like Kprobe, because data errors only occur in extreme cases.