Hi @Bing,
There isn’t a simple answer to this as it depends a bit on your data. The approach that I typically take is to start by setting --p-max-depth to a value around the median frequency per sample. In your case, this could be 25,000 (I’m just rounding to the nearest thousand from the information you provided, but rounding isn’t required). You may want to increase that value if the lines in the resulting rarefaction plot don’t appear to be “leveling out”, or decrease that value if you seem to be losing many of your samples due to total frequencies being closer to your minimum sampling depth than your maximum sampling depth.
I added these guidelines to the new section on Alpha Rarefaction Plotting in the Moving Pictures tutorial (added in QIIME 2 2017.10). There is some other important information in that new section, including definitions of some of the terms I’m using in my answer here, so be sure to read that.
Hope this helps!