Figure 1

Memory layout of SNPFile matrix. If your program only accesses a few columns at a time they will cluster nicely in virtual memory and it will be easy for the operating system to keep only the needed pages in physical memory. This means you can handle very big SNPFiles while not using very much actual memory. Furthermore, if your program only access columns ordered left-to-right and from top to bottom, the file will simply be accessed from the beginning to the end. This is what the entire computer, both hardware and software, is optimized for. Thus it should be very fast. If you read a row from the matrix, however, you will access a lot of pages in the file, only use a very small part of each and the operating system will waste a lot of time reading data that will not be used, since it operates on entire pages.