Too many backlog in DMA FIFO

[shuishen1983]

Hi I am using DMA FIFO for data aquisition. When I only have the data showing on the front panel, the data value is reasonable and the backlog value is around 3. But when I have the write to measurement file block in the block diagram, the back log value will rise to several thousands. How should I make the change to the program to reduce the backlog.

[BuddyHaun]

I looked at your code and you don't have the code you are using to write to disk on this block diagram. I would make sure that your loop time on the fpgaAcquire.vi is set to microseconds instead of ticks which it is currently set to. I would try the TDM Streaming vi's with the Open File and Close file outside the loop so that you are not opening and closing the file inside the loop. There are examples of TDM Streaming in example finder. Lastly if the TDM Streaming cannot be used due to file format restrictions or if it is still not fast enough you may need to create a second loop and use queue vi's to transfer the data. This will allow you to read multiple points from the queue using the flush queue function writing multiple points to the file at once. You will still probably need to open and close the file outside of the second loop. Look at some the Write Datalog File Example.vi to see how to do this.

[shuishen1983]

---

@Buddy Haun wrote:

I looked at your code and you don't have the code you are using to write to disk on this block diagram. I would make sure that your loop time on the fpgaAcquire.vi is set to microseconds instead of ticks which it is currently set to. I would try the TDM Streaming vi's with the Open File and Close file outside the loop so that you are not opening and closing the file inside the loop. There are examples of TDM Streaming in example finder. Lastly if the TDM Streaming cannot be used due to file format restrictions or if it is still not fast enough you may need to create a second loop and use queue vi's to transfer the data. This will allow you to read multiple points from the queue using the flush queue function writing multiple points to the file at once. You will still probably need to open and close the file outside of the second loop. Look at some the Write Datalog File Example.vi to see how to do this.

---

I did some experiment on the FPGA period. When I set it to be below 400 usec, the data looks almost the same. But when I set it to be above 500 usec. The period of the signal displayed on the screen seems to be multiplied by some constant, and the constant is not proportional to the FPGA period. I will agree with you on using TDMS streaming outside the loop and I will try it.

[nathand]

As I noted in this related thread, the timer is configured for Ticks, not uSec, which probably explains the timing issue.

[rex1030]

In general, write to disc operations are slow. If you are experiencing overflow as you seem to be describing, you need to get the write to spreadsheet file.vi out of the dma read loop. I recommend you use a producer-consumer architecture. This is where you have one loop reading values from the dma fifo (or any input in general) and putting the values into a queue. Then you have a second loop running in parallel that reads from the queue and writes the values to disc. It is better to write to disc in larger chunks (powers of 2 of course) in this consumer loop. Of course, it will take some tuning to find the right balance between queue size and write chunk size on your system, but I'm sure you can figure it out. Once you find that balance, I recommend doubling that queue size at minimum, to insure stability.