Nas with raid card spm393

15 4936
Edited by tkaiser at Mon Apr 27, 2015 14:04
mikronauts replied at Mon Apr 27, 2015 12:18
I'll try your script

On the RPi side all I fire up is simply this to get streams of 1 hour length (on the BPi the 'raw' h.264 streams get an MP4 container using GPAC/MP4Box so I decided to split the recordings into hourly pieces):
  1. raspivid -b 4000000 -w 960 -h 540 -t 3600000 -fps 30 -o - | nc -k -l 2222
Copy the Code
If you would do 1080p instead it should read "-b 16000000 -w 1920 -h 1080". Then the SoC will get a bit more hot. I don't believe this will ever be possible with BPi (as outlined in this thread)

Edited by tkaiser at Mon Apr 27, 2015 14:27
Maor replied at Mon Apr 27, 2015 13:01
Microserver => nonsence to me. It is a big overkill just for a network storage with the basic hdd failure protection.

Well, the problem with this sort of 'basic hdd failure protection' is that the feature you rely on (RAID) fails most likely when you need it. We do storage for a living and found that the least expensive solution that is able to provide data availability, security and integrity (three different beasts!) is something like a HP Microserver with a modern filesystem like ZFS/btrfs/ReFS.

Cheap NAS boxes suffer from the box itself being a single point of failure (how do you access your data if your proprietary RAID card has gone?) and unreliable operation. There are so many reports about the PSUs of drive enclosures failing when hot-plugging a replacement disk (second disk gets disconnected and all your data has gone) or rebuilds going wrong (especially the JMB393 used on your RAID adapter tends to data corruption when it gets too hot. Exactly the situation when a rebuild happens). I wouldn't spend money on additional devices that aren't able to protect from so many failures. But it's your data

BTW: Sorry for hijacking your thread with the camera stuff...

I made a first testing setup with the RAIDcontroller on my desktop and I am able to get 50-60MB/s write even when rebuilding.

@ tkaiser: do you have any link or more info about the data corruption? I hope, that it's not that crappy. I heared only good experience with that.
Anyway, I can put a heatsink on the controller.
Don't forget, that this is just a home network storage, I know, that for important data I need a decent backup. The only think is, when it fails, it will take some time to recover from the backup.

Edited by tkaiser at Mon May 4, 2015 02:26
Maor replied at Fri May 1, 2015 12:21
tkaiser: do you have any link or more info about the data corruption?

The JMB393 is the bigger sibling of the JMB321 and the only difference is the lack of the RAID processor in the latter (datasheets here and here).

In case you really do a stress test (utilizing eg. iozone or bonnie++) you will experience problems unless you can cool down the chip:

And while your drives have a 'good' URE rate of 10^-15 (please compare with and his older article explaining the problem linked from there) the whole idea of running a 4 disk RAID-5 with such a dumb RAID implementation (failing the whole array when a single remaining sector from another disk cannot be read after one HDD has gone) still makes no sense to me.

Instead of a crappy RAID-5 with 4 2TB disks I would go with one 6 TB drive on the SATA port and a second 8 TByte drive connected via USB. Then use mainline kernel, btrfs with "compress=lzo,noatime" options. Costs less and consumes less power. It's both faster and provides more storage space (due to transparent filesystem compression), provides more data integrity/safety due to btrfs checksums, regular snapshots that will be transferred incrementally to the second disk where you can then simply keep older versions of changed/deleted files. And even regarding data availability this is the better solution since when the 6 TB drive dies you can serve your data from the latest snapshot of the 8 TB disk. This is also some sort of a real backup solution since you keep versions on the larger disk and can restore from them in case you realize that data got deleted or damaged.

And you can have an eye on drive health using SMART (checking the value of the common attributes and triggering SMART self tests on a regular basis). For your purpose the relatively cheap Seagate Archive v2 HDDs would match. And if you use a SAT capable USB-to-SATA-bridge that works with smartmontools out of the box then you could even use SMART with the USB connected disk (even my el cheapo Logilink AU0006D containing the JM20337 -- vendor/device ID 152d:2338 -- provides that feature which is an absolute prerequisit to use HDDs for me). Not being able to monitor drive health is a no-go criteria for any storage solution.

Edited by Maor at Mon May 4, 2015 03:30

You made a good point. The option with one large HDD is worth considering, but no way of going with seagate, seagate has the highest failure rate, good would be Hitachi, but their desktop drives are no longer available :-(. For now, I will give a chance to JMB.

Edited by tkaiser at Mon May 4, 2015 05:40
Maor replied at Mon May 4, 2015 03:27
For now, I will give a chance to JMB

Good luck!

Just in case: I would have a look here if you once get in trouble after a disk failure: You'll need a PC with at least 3 free SATA ports and appropriate disk space somewhere else to copy over the whole contents of your array (another huge disadvantage of 'dumb' RAID happening at the block level compared to smarter RAID attempts like in zraid oder btrfs' implementation).

Regarding Seagate: according to Backblaze especially the 3TB and 1.5TB Barracudas make problems:

You have to log in before you can reply Login | Sign Up

Points Rules