Table of Contents
Samba Tuning
| Reference | Samba tunning | mmap and DMA |
Main Technique
CONFIG_NET_DMA to offload TCP copy from kernel to userspace ??
zero copy
kernel bypass
min receivefile size
This option changes the behavior of Samba when processing write calls. Any incoming write call on a non-signed SMB/CIFS connection greater than this value will not be processed in the normal way but will be passed to any underlying kernel recvfile or splice system call (if there is no such call Samba will emulate in user space). This allows zero-copy writes directly from network socket buffers into the filesystem buffer cache, if available. It may improve performance but user testing is recommended. If set to zero Samba processes write calls in the normal way. To enable POSIX large write support (SMB/CIFS writes up to 16Mb) this option must be nonzero. The maximum value is 128k. Values greater than 128k will be silently set to 128k.
use sendfile
If this parameter is yes, and the sendfile() system call is supported by the underlying operating system, then some SMB read calls (mainly ReadAndX and ReadRaw) will use the more efficient sendfile system call for files that are exclusively oplocked. This may make more efficient use of the system CPU's and cause Samba to be faster.