Here are some notes that are still issues in the svn version of March 27th. This does not include updates to the bibliography, which I have sent to Rainer separately. Bill Contents (and forward) - The capitalization of section headings is inconsistent. This is apparent when looking at the contents. p29. The tables of types in C do not include the types added in MPI 2. Either they should be added or there should be a forward reference to them. p59, line 13, example 3.15. The MPI_Recv is missing a "status" argument. p104, line 4. MPI_GET_ELEMENT should be MPI_GET_ELEMENTS p106, line 13, section 3.12,14, the MPI_COMM_RANK call is missing the ierr argument. Line continuation is also not correct for Fortran. p106, line 40, section 3.12.14, the MPI_COMM_RANK call is missing the ierr argurment. Line continuation is also not correct for Fortran. p107, line 15, section 3.12.14, the MPI_COMM_RANK call is missing the ierr argument. Line continuation is also not correct for Fortran. p107, line 27, section 3.12.14, the MPI_COMM_RANK call is missing the ierr argument. Line continuation is also not correct for Fortran. p133, line 46 (MPI_GATHERV) the array index should be [j], not [i], to match the following text. p134, line 44, the myrank argument of MPI_Comm_rank needs to be &myrank. p149, line 41, the Recv is missing the recvtype before the "i". p155, lines 39-end of page - the table of predefined reduction operations for C needs to be reformatted. p160, line 13, example 4.18, remove the semicolon in the Fortran declaration of err. p160, line 15, example 4.18, the MPI_COMM_RANK call is missing the ierr argument. p160, line 21-21, example 4.18, the REDUCE call is missing the "CALL" and has an (invalid) semicolon at the end. p178, line 28, "in not" should be "is not" p179, lines 44-47. This text, about processes dynamically joining an MPI execution, dates from MPI-1 and is misleading. This entire paragraph needs to be replaced. p187, lines 5-8. This advice to implementors on reference counts for groups should include MPI_COMM_GROUP as a routine that increments the reference count. p197, line 8, section 5.5.1, the status argument is missing from MPI_Recv p198, line 23, Section 5.5.3. The closing brace on the test if (me ! = 0) should be moved after the MPI_COMM_FREE, other MPI_COMM_FREE will be called with MPI_COMM_NULL, which is erroneous. p198. Section 5.5.3, there is an extra space in "( commslave)" p194, Figure 5.2 has the left edge of each box clipped off p199 Section 5.5.4, example 4: t&hecomm -> &thecomm p199: the close } on the "me != MPI_UNDEFINED needs to be moved to after the MPI_Comm_free, otherwise thecomm is MPI_COMM_NULL. p200, line 45. The value of the status point must be initialized. For example, use MPI_Status status, and pass the address of that status to the two MPI_Wait calls. p208, top of page. Figure 5.3 has the left edge of the left square clipped off p210, top of page. Figure 5.4 has the left line clipped off. William Gropp Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign *