PCI riser cards

[LordMorpheus]

I work with an autonomous vehicle research group. Our robot is a Chevrolet Tahoe with (among other things) a rack containing 16 1u computers in the back. Several of these have various DAQ cards and other devices installed, and these machines have been points-of-failure for the robot. Either the DAQ card won't detect correctly or the machine won't even boot. So far, re-seating the DAQ card and the PCI riser used to install it has fixed the problem.

I'm aware that PCI risers are often more trouble than they are worth, but I am wondering if there exists a "good" PCI riser card that is better/more reliable than what we currently have installed.

I don't know what brand/model risers are currently used, but they are RCP122 cards much like this one: http://www.worldnetcomputers.com/def/catalog/details.aspx?categoryId=1117&itemId=14938

To sum up, I know some of you have experience with servers, and that some of you must have experienced the headaches that riser cards can cause. Is there a "good" riser we can buy to avoid or at least reduce the frequency of problems?

Thanks!

edit: Looks like the riser might be case-specific? This is looking like a long shot. Maybe a thorough cleaning of the contact points would help?

 

[paperwastage]

what exactly is the problem? too much rough movement on the car, causing the PCI card to misalign/pop out from the riser?

 

[mfenn]

16 1U boxes? Why are you doing that vs. a couple of 4Us that are more powerful, have more aggregate PCI(e) slots, and don't have to bother with risers?

 

[LordMorpheus]

what exactly is the problem? too much rough movement on the car, causing the PCI card to misalign/pop out from the riser?

The risers aren't coming out (the rack itself is mounted on vibration isolators), we're just losing the ability to talk to the cards plugged into them. Re-seating the cards/risers fixes the problem, so I suspect the issue has to do with the riser (I had an HTPC box years ago that used an AGP riser to fit the graphics card, and it was constantly refusing to boot. Re-seating the card and riser would "fix" this issue until it happened again. It was my understanding the risers are generally a pain in the ass). Everything looks fine mechanically when I open the box.

16 1U boxes? Why are you doing that vs. a couple of 4Us that are more powerful, have more aggregate PCI(e) slots, and don't have to bother with risers?

The car was built before I joined the group, and the group that did build it has graduated, so I'm not entirely sure. I think the idea was to sidestep the issue of running multiple time-sensitive pieces of software on the same machine - none of them are running real-time operating systems.

The various algorithmic components that make the car what it is need to be run more-or-less in real time, and running two demanding programs on the same machine (that isn't running a real-time OS) causes unpredictable behavior regarding what gets executed when. Of course, careful development (or a true real time implementation) makes this a non-issue, but given the time frame the car was built in I think it was just easier to try to have one computer per software component - one computer integrates GPS, IMU, and several other information sources to estimate the vehicle's pose, one computer runs the high level logic, one computer runs the low-level path planning, lots of computers perform object detection for the various sensors on the vehicle, &c. Only three of these computers need expansion cards, as all but three of the sensors on the car communicate with the network through dedicated microcontrollers.

 

[janas19]

Off topic, but this is deliciously interesting - is the server connected to the vehicle for function, or simply mobility?

 

[mfenn]

The car was built before I joined the group, and the group that did build it has graduated, so I'm not entirely sure. I think the idea was to sidestep the issue of running multiple time-sensitive pieces of software on the same machine - none of them are running real-time operating systems.

That is understandable, but risers are very motherboard and case specific, so I don't think that you'll have much luck finding one that performs significantly different from what you have now. Hence the suggestion to rebuild with a different form factor.

 

[LordMorpheus]

Off topic, but this is deliciously interesting - is the server connected to the vehicle for function, or simply mobility?

You could in theory do your processing offboard but you would need a very good wireless link between the car and your computing resources. You see stuff like this when people run intensive navigation or mapping algorithms on little robots like a quadrotor -- they'll follow the robot with a laptop that does the heavy computational lifting. For the car the ranges are greater making wireless less attractive, and there's no size/weight reasons not to do the computation on board. The computers in the car all run different parts of the software that takes information from sensors and turns it into actuation commands on the vehicle.

That is understandable, but risers are very motherboard and case specific, so I don't think that you'll have much luck finding one that performs significantly different from what you have now. Hence the suggestion to rebuild with a different form factor.

I was hoping they were more standardized, but looks like what we have is what we have. I might pick a few spares up anyway, before they are impossible to find. Thanks for the help.

 

[dave_the_nerd]

Old computers used to have a lot more screws holding shit together.

</grumpyoldman>