We have been trying to troubleshoot a problem and are at a dead end.
Using ROS 4.11, 4.16 or 4.17 same results
We have four different RB433UAH boards. Two have 64MB of NAND and two have 512MB of NAND. MT only promises to build boards with 64MB so sometimes we get boards with 64 and sometimes we get 512. Sometimes we even get 1GB, it all depends on the weather in China (or the availability and cost of NAND flash at the time of build).
Anyway, we have two SR9 modules and run them using Nstreme in a 10MHz channel. When we run two boards with 512MB of NAND, we get great performance and throughput. Generally in the 10-12Mbps range. Then when we use two 64MB of NAND boards we get crap throughput of less than 100kbps. Signal strength goes down too. If we use one 512MB and one 64MB we get crap performance too.
We have tested every possible configuration. We have even run test to an ALIX 3D board with 512MB CF and 64MB CF. We found that the ALIX worked fine with either CF to a RB433UAH with 512MB and worked fine from ALIX to ALIX, but when we used a RB433UAH with 64MB it crapped out.
We have eliminated all other possible conflicts. It has to be related to NAND size. But we really don’t get why. It makes no sense.
So we have a customer that is now demanding that we only supply RB433UAH with 512MB of NAND or more. But since we cannot order NAND size on the boards, there is no way for us to know what NAND we will get when we order. Most of our recent boards have all been 64MB.
I posted earlier that I really wish MT would state a spec on NAND size and stick with it. But no reply.
Any ideas of how to fix this? We are out of options.
More testing shows that there is absolute consistency in having poor performance with 64MB NAND. We know that this is not the issue, but it does point specifically to builds of RB433UAH boards.
What we know with certainty is that the exact same SR9 modules, with the exact same antennas, and the exact same router configuration (copied using export from one board to the next) shows that boards with various serial numbers (build dates) with 64MB NAND show good signal, good snr, good CCQ until a load is applied (traffic) and then the CCQ goes to crap and the link has terrible performance. Putting exact same SR9 modules in RB433UAH with 512MB NAND from various build lots we see perfect performance. So while the problem may not be NAND size, it is related to the hardware. All other items are same.
We did load 4.17 with the NV2 wireless driver. We changed from using Nstreme to using NV2. In this case, we still have same exact issue with Nstreme. If we switch over to using NV2, then the problem changes from no link to a link that drops every few seconds. Again, we see no drops with 512MB NAND boards. I would say given the complete lack of real diagnostics available in MT regarding kernel and drivers, that it might be some type of buffering issue. Like a memory buffer is filling up and overflowing. There are no debug errors, no wireless errors, nothing.
Is there anyway to assure our customer that we can deliver 512MB NAND boards? While this doesn’t correct the problem, it does eliminate the symptoms. Otherwise we will have to look at moving to ALIX boards which do not exhibit this problem. We really don’t want to go backwards like this. There are so many other issues that the Geode design with flash had.