to sendfile(2) or not to sendfile(2)

Poul-Henning Kamp phk at
Wed Feb 1 11:26:01 CET 2012

Since the earliest days, Varnish have had support for sendfile(2) on
systems where sendfile(2) worked well enough.

That is roughly FreeBSD 8.x and later.

There are theoretical reasons why sendfile(2) should be faster than
writev(2), provided you use a file based storage (-sfile, -spersistent)
but I have yet to see anybody telling me that these theoretical speedups
matter, or are even visible, in reality.

If you have the "sendfile_threshold" on your varnishd, you should have
a working sendfile, could you please try to change it contents and
tell me if you can see _any_ difference at all ?

VM parameters like page-faults per second etc might be where to look
for changes if you cannot see them elsewhere.

Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

More information about the varnish-misc mailing list