[Scalapack] Communication complexity on PBLAS routines

[Ken Habgood]

I was doing some research on parallel algorithms for linear algebra and was 
curious about how communication complexity?is typically?measured.

Is there a good paper that evaluates the communication complexity or 
communication requirements of the PBLAS routines?? I've found a number of 
papers that discuss the computational compleixty and overall efficiency of 
SCALAPACK and PBLAS but I was hoping there was some literature that gave a 
detailed?discussion of?the communication requirements of the algorithms.? 

I'm mainly focused on LU-decomposition so the PDGETF2, PDTRSM and PDGEMM 
routines.

Thanks
Ken Habgood