GLM is slow on large datasets. Using OnlineStats for regressions? MixedModels?

[pdeffebach]

(X'X)\(X'y) is the canonical way that OLS is estimated.

But what are you trying to do? In my experience, linear regression without standard errors is not very useful.