SSD drive corruption with LabView real-time?

[dmccarty]

Thanks everyone.  It sounds like in our goal to make the disk more reliable (heat was an issue back then as well) we might have made things worse because of the poor SSD support.  So even though we're doing a vision app with NI real-time, would it make more sense to use a high-RPM spinning disk HD instead?

[rolfk]

SSD for temporary storage of high speed vision data would never be an ideal solution, up to today. SSD wear is real and not a fairy tale, even in modern high end SSDs although the technology gets increasingly more long term reliable. It won't really matter for normal workstations nowadays, but we did for instance have a few years ago an issue where our notebooks were delivered with second grade SSD disks because that saved the seller a few bucks. They all consistently started to fail within 3 to 6 months after purchase. We ended up having them replaced by Samsung drives which haven't shown any failures so far. But even those would not quickly been used in data centers as they would wear pretty quickly and cost still considerably more. When you are needing Peta Bytes of storage $/TB is a real factor to account for, but lifetime of the drive is also something to consider.

While a HDD for data center use will typically have a life time of 8 to 10 years, almost irrespective of how many times it is written to or read from, SSDs will not simply fail after a certain amount of time, but after a certain amount of write accesses to a single storage cell. For data centers with its often very frequently changing data patterns this quickly can become a consideration.

[rolfk]

Just one other thought. It's not something to just go and do at this moment but if you ever plan to do some redesign on this system (hardware availibility can be a bitch too nowadays, with most PC hardware being considered obsolete by the manufacturers the moment they hit the shelfs in the stores, and industrial grade hardware often only being slightly better and usually at least two generations behind the current technology 😀).

Since NI is about to discontinue Pharlap ETS for their products soon, you also want to get rid of that as soon as you make any significant changes to the current system. While not (yet) officially supported by NI like the RTPC setup that you have now, there is a possibility to use NI Linux RT on non NI hardware. The legal licensing issues is still something you would have to take up with NI at this point. They haven't yet released official statements about this other than that you do need a valid license from NI for this and an email address to contact for questions about this. If you can convince NI that you have a business case here, they might be willing to negotiate some licensing with you.

The technical challenges to do this are not trivial either, especially if you also want to incorporate NI Vision and especially hardware, like frame grabbers and or DAQmx. To get the bare NI Linux RT running is however fairly easy and just needs some intermediate Linux knowledge really. If you don't know what a Unix shell or a shell script is, you definitely won't succeed. And being able to use the arcane vi text editor is also a bonus as you will almost certainly have to make a few changes to some startup scripts to make it match with your hardware setup.

Getting things like IMAQ Vision installed shouldn't be impossible either. Hardware drivers like IMAQdx or DAQmx on the other hand will likely proof very challenging. I wouldn't try that without some committment from NI to support this actively with product specialists. In fact I think incorporating non-NI hardware (that would include things like GigE cameras which technically involve no extra hardware from NI) would be a lot easier to do although quite a bit of work as you would need to write the LabVIEW interfaces to that yourself.

[rtollert]

There are a couple of things I can suggest.

First, you didn’t mention what filesystem the disk is formatted as. It really, really needs to be formatted Reliance.

Second, you didn’t mention what your power situation is. In general, if you can’t preclude power loss within 5-30 seconds after the last write to disk, you have a slightly elevated risk of being exposed to SSD firmware bugs involving power recovery. (Don’t ask me how I know this.)

[dmccarty]

@rolfk thanks for all the tips and advice.  We have, since this system was delivered, redesigned the hardware from the ground up and are now using a commercial IPC running Windows and running LV on that.  We're still using the NI frame grabber/FPGA so in that sense we're tied to NI hardware, but so far (knock on wood) the systems have been rock solid.

@rtollert yes I'm almost positive the systems are running Reliance FS.  Unfortunately there's a power switch on the outside of the unit (or on the entire in the plant that this runs in) and we can't prevent the customer from shutting off power whenever they want.