[ghorrocks]

Is ITW DNS? No, it is not. ITW is how a turbulence model deals with the wall boundary condition. DNS does not use a turbulence model at all as all turbulence scales are resolved. But ITW is more expensive than wall functions.

Is ITW too costly for real CFD applications? That depends on what you are modelling and what resources you have available. There are certainly many cases where it is practical.

Is ITW more accurate than AWT? Sometimes, but not all the time.

[Far]

Regarding the ITW, is that kind of large discretization error DNS method that is usually too costly to be used in the real CFD application? For boundary layer separation predictions, will ITW be theoretically much more accurate than AWT?

ITW is a RANS approach for solving the viscous sub layer. ITW is not always needed, however if the yplus is less than 10, both ITW and advance wall functions (aka hybrid wall function) will give you the same results.