[linux-cifs-client] Re: Asynchronous readahead limit

Steven French sfrench at us.ibm.com
Sat Apr 9 01:59:26 GMT 2005






One other factor that should make a difference is the extra copy going on
in the read path and write path.    When we get rid of that copy, and
change to small buffers in cifs_demultiplex_thread for all SMB responses
under 400 bytes or so that should cut the load on the page manager a lot
(we are beating the page manager pretty hard with large allocs - in some
cases for quite small requests and responses)

It is also going to be interesting to test with and without
CONFIG_CIFS_EXPERIMENTAL enabled since I added the calls to the generic
readv/writev/aio calls recently when CONFIG_CIFS_EXPERIMENTAL is enabled.

This is going to be fun to figure out - I am fairly certain we can beat ftp
pretty easily by the time we are done :)

Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench at-sign us dot ibm dot com
-------------- next part --------------
HTML attachment scrubbed and removed


More information about the linux-cifs-client mailing list