Depends on what you want to know. If you look at a sample of 100K with an actual winrate of 1BB/100 and a SD of 17BB/100, you'll see there is a relatively small chance that your result will be negative after a 100K sample.
Sure a 95% confidence interval is still roughly between 0-2BB/100, so for reliability on your 'exact' winrate (which is undermineable anyway as your own play as well as your opponents change over such a long time) you'll certainly need to have a much larger sample.
But 100K is a starting point for basics.
And oh well. I run -1.5BB/100 over a 100K database on $2/4 myself and I'm still not really convinced that I play that bad so...