The FILE_FLAG_NO_BUFFERING takes this concept one step further and eliminates all read-ahead file buffering and disk caching as well, so that all reads are guaranteed to come from the file and not from any system buffer or disk cache. When using FILE_FLAG_NO_BUFFERING, disk reads and writes must be done on sector boundaries, and buffer addresses must be aligned on disk sector boundaries in memory.
These restrictions are necessary because the buffer that you pass to the read or write API is used directly for I/O at the device level; at that level, your buffer addresses and sector sizes must satisfy any processor and media alignment restrictions of the hardware you are running on.
This code fragment demonstrates how to sector-align data in a buffer and pass it to CreateFile():
char buf[2 * SECTOR_SIZE - 1], *p;
p = (char *) ((DWORD) (buf + SECTOR_SIZE - 1) & ~(SECTOR_SIZE - 1));
h = CreateFile(argv, GENERIC_READ | GENERIC_WRITE,
FILE_SHARE_READ | FILE_SHARE_WRITE, NULL, CREATE_ALWAYS,
FILE_ATTRIBUTE_NORMAL | FILE_FLAG_NO_BUFFERING, NULL);
WriteFile(h, p, SECTOR_SIZE, &dwWritten, NULL);
If you have a situation where you want to flush all open files on the current logical drive, this can be done by:
hFile = CreateFile("\\\\.\\c:", ....);
When opening a remote file over the network, the server always caches and ignores the no buffering flag specified by the client. This is by design. The redirector and server cannot properly implement the full semantics of FILE_FLAG_NO_BUFFERING over the network. In particular, the requirement for sector-sized, sector-aligned I/O cannot be met. Therefore, when a Win32- based application asks for FILE_FLAG_NO_BUFFERING, the redirector and server treat this as a request for FILE_FLAG_WRITE_THROUGH. The file is not cached at the client, writes go directly to the server and to the disk on the server, and the read/write sizes on the network are exactly what the application asks for. However, the file is cached on the server.
Not caching the client can have a different effect, depending on the type of I/O. You eliminate the cache hits or read ahead, but you also may reduce the size of transmits and receives. In general, for sequential I/O, it is a good idea to cache on the client. For small, random access I/O, it is often best not to cache.