external disk enclosures meet high-volume writes

Have you wondered about the differences between all those external USB-to-SATA enclosures, that pile up in your drawer/side board/whereever? Of course there are the typical differences in size, color, robustness and USB2 vs. USB3 – but we recently stumbled across an unexpected difference: Mass data reliability.

To shorten recovery time after a complete storage crash for a small office environment, we looked into weekly “to disk” backups in addition to the tape-based backup solution that is already in place. That’ll be less than 1 Terabyte of data per week, and we wanted some automated way of creating the backup and support for taking the backup disk off-site after the backup. Using two alternating disks would ease that task: In week 1, the backup is written to disk A, which is then replaced by the admin using disk B. Disk A is taken off-site, and on the next backup cycle, disk B is written. The admin replaces disk B by disk A, takes disk B off-site and so on.

With hot-plugable USB disks, this sounded easy enough. We had a bunch of new 2,5 1TB disks at hand, and going for the max, we decided to try a RAID0/1 USB3 enclosure with room for two disks… the idea was to split the RAID set and take one disk to a safe place and leave one disk in the office… but we never came that far. During our tests, writing the then 500 GB of backup data (several large tar files and some 20 GB+ virtual disk images) failed with a disk error. The error didn’t happen at once, but took it’s time (it happened at around 400 GB written).

Although these were brand-new disks, such problems do occur… but when the third disk (out of a 10-disk shipment) died, I became more than curious. Mounting the “defective” disks in a testing server showed no SMART reports of any problem and the disks could be read and written without trouble.

The enclosure was not only freshly bought, but new to us in it’s RAID chip set, so we decided to try a simple (USB2) enclosure, grabbing one from the shelf. While at first things seemed to work out, a similar error happened again, after a few backup cycles. So it’s the disks, after all? All the enclosures have been used some time and were trusted, as no obvious errors had been observed until now. But: The typical usage pattern wasn’t “writing long streams of continuous data for hours”…

Trying out more combinations, we by chance picked one of our most solid enclosure models (“Icy Box”) with a JMicron chipset (“JM20329 SATA Bridge” with USB ID 152d:2329). This seems to be our lucky number: using the same disks, we have yet to see any reports of disk errors, even after 6 months of operation.

So what we learned is that while many USB-to-SATA bridges seem to work nicely in typical day-to-day operation, you may run into problems when feeding continuous high-volume data streams through them.

I will not make any brand name recommendation here, on purpose, because even with the same product name, chip sets and more may change. So on one hand, you’ll have to test yourself if what you have suits your needs – and on the other hand, you should not rule out the enclosure as a potential cause of errors.

This entry was posted in Linux. Bookmark the permalink.

Leave a Reply