How precise is realized volatility?

A stock's log price is modeled as $X_t = \mu t + \sigma W_t$ with $\sigma = 20\%$ and $\mu = 10\%$ a year. You estimate $\sigma^2$ from one year of data with the realized variance

over $n$ equal steps.

(a) Take $\mu = 0$ first. Find $E[RV]$ and the standard deviation of $RV$ for daily data, $n = 252$. Roughly how many volatility points of error does that give for $\sigma$?

(b) How many steps do you need for the relative standard error of $RV$ to be 1%? About how many 5-minute bars per trading day is that?

(c) Now put the drift back. How much does it bias $RV$ at $n = 252$? Can more frequent sampling pin down $\mu$ as well?