A SDR peripheral like the Ettus USRP B210 continuously digitizes the incoming RF and sends it to the attached computer. There are no gaps in its coverage — anything that is within the bandwidth is captured in full.
If you use one to create a spectrum analyzer, the minimum duration of signal you will be able to observe will be determined by the algorithms used by the software that you use to process the signal; on a spectrum plot (power vs frequency) in particular, the windowing and degree of overlap between successive FFTs of the incoming signal. GNU Radio's “Frequency Sink” is very poor at visualizing short signals, since it does not overlap at all (so a pulse could land at the end of one FFT-window and the beginning of the next, or squarely in the middle of the window, and show up very differently or not at all), unless someone's improved that since the last time I checked.
On the other hand, you could write a program that looked for single samples with significantly higher amplitude, and then analyzed the region before and after it. This would consistently capture arbitrarily short pulse signals (up to the limit where a short pulse cannot be considered to be a narrow-band signal), and since it is analyzing a snapshot, it doesn't have to run in “real time”.
Again: It's all up to the software. That's the point of software-defined radio. Unfortunately, I don't have the math handy to tell you how to quantitatively measure the capabilities of a given sof, but I can assure you that the hardware is irrelevant — every SDR receiver works this way.