Best Storage Solution?

[JoeComp]

In theory, bitrot shouldn't matter during media playback, but in practice, it does.

False.

Bit-rot is totally irrelevant for media playback. Bit-rot (aka silent corruption) refers to a very, very, very unlikely event where one or more stored bits changes without the storage device signaling an error. With current HDDs, these events are less likely than UREs (unrecoverable read errors), which are already 10^-14 or 10^-15 events.

A faulty SATA controller munging data is not bit-rot. And while filesystem checksumming would be able to catch that sort of error, there are other possibilities for errors in the stream from media source to media player that would be outside the realm of filesystem checksumming. Hence, you do not actually have "end to end checksumming" in the context of obtaining and playing back media.

 

[Rectal Prolapse]

Whoops I mean corruption not bit-rot. Wherever the corruption may be.

At least the original video archive was on a different machine without the corruption (offline).

 

[iroc409]

With current HDDs, these events are less likely than UREs (unrecoverable read errors), which are already 10^-14 or 10^-15 events.

There were some interesting articles posted a while back on here that indicated with larger drives with a 10^14 URE rate, a data failure is almost guaranteed. Hence why RAID5 has pretty much become obsolete. Some of the articles indicated a standard mirror would also no longer protect the system, that a 3-way mirror would be required to prevent corruption.

I don't remember the math off the top of my head, but I think it's when you get over 4TB single drives.

 

[JoeComp]

There were some interesting articles posted a while back on here that indicated with larger drives with a 10^14 URE rate, a data failure is almost guaranteed. Hence why RAID5 has pretty much become obsolete. Some of the articles indicated a standard mirror would also no longer protect the system, that a 3-way mirror would be required to prevent corruption.

As I said, silent corruption is even less likely than a URE event on current HDDs.

And UREs are very rare and not an issue worth worrying about for media playback, only for RAID rebuilds. All of the solutions discussed here, except for pooling-only or UnRAID, support dual-parity or higher.

 

[_Gea]

About undetected silent errors

The only serious study about silent data errors I am aware of at the moment is the study from CERN that claims a error rate of 3 x 10 -7 failures what means that you must espect 30 undetected errors on a 10 TB array. With media files this may be a red dot instead a blue one, with some files this may be a whole file loss, with financial data this may be a -100000 instead a +100000.

http://indico.cern.ch/event/13797/session/0/material/paper/1?contribId=3
http://storage-news.com/2009/09/29/have-you-ever-worried-about-silent-errors/

This study is from 2007. You can speculate if modern disks are better due to a better error correction or worser due to a much higher density. I would espect the second.

 

[JoeComp]

The only serious study about silent data errors I am aware of at the moment is the study from CERN that claims a error rate of 3 x 10 -7 failures...

That is not bit-rot of data correctly written to an HDD and then read back incorrectly without a reported error by the HDD.

The CERN study is including all sorts of errors, which could include RAID card issues, power supply glitches, transmission errors, etc.

Since the silent corruption error rate of data sitting on an HDD is obviously less than the URE rate, which is less than 10^-14, the obvious takeaway from the CERN study is that bit-rot is a minor concern relative to errors due to faulty hardware.

And that leads invariably to the conclusion that ANY hardware that is not part of the filesystem chain could also be a major source of data errors, and thus when considering digital media playback from most fileservers there are plenty of potential errors sources that would NOT be caught by filesystem checksumming. That is why "end-to-end error protection" is a highly misleading term to use when discussing playback of digital media from a fileserver. I suggest the more accurate and descriptive terms "filesystem checksumming" or "filesystem hashing".

 

[Rectal Prolapse]

I don't think gea was talking about bit-rot in his post. I am sure he was aware of that.

"undetected silent errors" <-- that was what he is talking about.

UREs (I assume that means Unrecoverable Errors) are a big pain in the ass for media files - when an entire sector is messed up you could lose more than just a pixel - it could break the stream completely and then your player stutters or pauses as the media tries to figure out what it should do with the error. On optical discs it usually has a big pause as it tries to reread the sector (at least, that is what I *think* is happening), followed by picture break up.

I find that UREs in media is far more common than "silent" bit-rot corruption. It's the big reason why I rip my discs onto hard drives - a spec of dust on a blu-ray can halt playback completely, and I don't feel like polishing them every time I want to watch a movie. It's dry where I live so dust gets everywhere!

I've also had discs fail - the dye appears to erode - it actually looks like it is rotting when you look at it with a loupe. *sigh*

 

[JoeComp]

UREs (I assume that means Unrecoverable Errors) are a big pain in the ass for media files....

False.

UREs are a minuscule concern with playback of media files. The URE bit error rate is less than 10^-14 for current consumer HDDs. That is an average of about 1 bit error for every 12TB of data (and that is worst case spec, in reality UREs are probably less likely). So, if you play back 12TB of media files, which if we are talking full-quality blu-ray rips, that is about 800 hours of media, you will average one 512 byte sector being unreadable. The chance that one 512 byte sector being unreadable will cause a major problem with media file playback is small. So, we are talking about a small chance of one major media playback problem out of every 800 hours of media playback (if you watch one movie every day, that is more than a year of movies). And that is worst case -- most people will probably have an even smaller chance of problems than that already tiny chance.

As I said before, for a home media fileserver, UREs are only a concern during RAID rebuild. Anyone with more than about 12TB of data should be running at least dual-parity to minimize problems during rebuilds. However, with SnapRAID, even having a URE too much during a rebuild is not the end of the world, since SnapRAID will still be able to restore the majority of your data, with usually only one file being unrecoverable due to the URE.

 

[brutalizer]

False.

UREs are a minuscule concern with playback of media files. The URE bit error rate is less than 10^-14 for current consumer HDDs. That is an average of about 1 bit error for every 12TB of data (and that is worst case spec, in reality UREs are probably less likely).

This is false. If you bothered to follow the data storage scene, you would have known that in reality the error rates are much higher than 10^&#8315;14, because of other external factors than the hard disk itself. There might be errors in the PSU, SATA cable might be slightly loose, bugs in disk firmware, loud sound causes vibrations, etc. Instead of just repeating the same false things over and over again, I suggest - again - that you read the research papers on data corruption. I also post some research from CERN (as _Gea did) that shows the error rates are far higher than the disks URE. There are many other factors that can go wrong. Why do you repeatedly ignore all the data corruption research and write false things?
http://en.wikipedia.org/wiki/Silent_data_corruption#SILENT

"....There are many error sources beyond the disk storage subsystem itself. For instance, cables might be slightly loose, the power supply might be unreliable,[2] external vibrations such as a loud sound,[3] the network might introduce undetected corruption,[4] cosmic radiation and many other causes of soft memory errors, etc. In 39,000 storage systems that were analyzed, firmware bugs accounted for 5&#8211;10% of storage failures.[5] All in all, the error rates as observed by a CERN study on silent corruption are far higher than one in every 10^-16 bits.[6] Webshop Amazon.com confirms these high data corruption rates.[7]..."

 

[JoeComp]

This is false.

Wrong. I was correctly discussing UREs, which are a specific type of error.

And the research you cite is not relevant to a home media fileserver.