--- ioscale.tex.orig	2004-06-10 13:55:29.344268986 -0700
+++ ioscale.tex	2004-06-10 13:59:11.755156683 -0700
@@ -138,10 +138,15 @@
 
 \subsection*{Local allocation of consistent DMA mappings}
 
-Consistent DMA mappings are used frequently by drivers to store command
-and status buffers.  They are frequently read and written by the device
-that owns them, so making sure they can be accessed quickly is
-important.
+Consistent DMA mappings are used frequently by drivers to store
+command and status buffers.  They are frequently read and written by
+the device that owns them, so making sure they can be accessed quickly
+is important.  The table below shows the difference in the number of
+operations per second that can be achieved using local vs. remote
+allocation of consistent DMA buffers.  Local allocations were
+guaranteed by changing the \emph{pci_alloc_consistent} function so
+that it calls \emph{alloc_pages_node} using the node closest to the
+PCI device in question.
 
 \begin{table}[h]
 \begin{tabular}{|l|l|}
@@ -155,6 +160,6 @@
 \caption{Local vs. remote DMA buffer allocation}
 \end{table}
 
-This is a relatively simple change that platforms can make in their
-consistent DMA allocation routines, using the Linux NUMA topology API
-(though that API may need some changes to make it easy).
+Although this change is platform specific, it can be made generic if a
+\emph{pci_to_node} or \emph{pci_to_nodemask} routine is added to the
+Linux topology API.