MFC r275568:

Count consecutive read requests as blocking in CTL for files and ZVOLs.

Technically read requests can be executed in any order or simultaneously
since they are not changing any data.  But ZFS prefetcher goes crasy when
it receives consecutive requests from different threads.  Since prefetcher
works on level of separate blocks, instead of two consecutive 128K requests
it may receive 32 8K requests in mixed order.

This patch is more workaround then a real fix, and it does not fix all of
prefetcher problems, but it improves sequential read speed by 3-4x times
in some configurations.  On the other side it may hurt performance if
some backing store has no prefetch, that is why it is disabled by default
for raw devices.

1 files changed, 8 insertions, 1 deletions

diff --git a/usr.sbin/ctladm/ctladm.8 b/usr.sbin/ctladm/ctladm.8
index 21b1946..e23e743 100644
--- a/usr.sbin/ctladm/ctladm.8
+++ b/usr.sbin/ctladm/ctladm.8
@@ -34,7 +34,7 @@
 .\" $Id: //depot/users/kenm/FreeBSD-test2/usr.sbin/ctladm/ctladm.8#3 $
 .\" $FreeBSD$
 .\"
-.Dd November 5, 2014
+.Dd December 6, 2014
 .Dt CTLADM 8
 .Os
 .Sh NAME
@@ -995,6 +995,13 @@ command sequence order shall be explicitly handled by the application
 client through the selection of appropriate commands and task attributes.
 The default value is "restricted".  It improves data integrity, but may
 introduce some additional delays.
+.It Va serseq
+Set to "on" to serialize conseсutive reads/writes.
+Set to "read" to serialize conseсutive reads.
+Set to "off" to allow them be issued in parallel.
+Parallel issue of consecutive operations may confuse logic of the
+backing file system, hurting performance; but it may improve performance
+of backing stores without prefetch/write-back.
 .It Va rpm
 Specifies medium rotation rate of the device: 0 -- not reported,
 1 -- non-rotating (SSD), >1024 -- value in revolutions per minute.