questions on ARMA and KPSS

Hi,

I have been fitting a series of data representing 
a week of Internet traffic (which is daily seasonal 
and have a general trend toward lower rate at the 
weekends). 

Before I do the ARMA fit (which takes care of
seasonality with a lag equal to one day), do I 
have to make sure the data is stationary? From the
results and visually, it seems that this was taken 
care of. But the residual failed the kpss test with 
a p-value less than 0.01.

Thank you,
Weiguang