On Sat, Apr 06, 2002 at 08:53:06PM -0500, Sean O'Connell wrote:
> Hi-
>
> I have opened a bugzilla request with the redhat regarding applying some of
> the recommendations from the nfs how-to performance section (chapter 5) and
> also adds a configuration file that lives in /etc/sysconfig and is read in
> by the startup script /etc/init.d/nfs. This was based on a series of con-
> versations with Seth Vidal regarding ways of incorporating the performance
> recommendations for our servers.
>
> The changes are detailed in the following bugzilla report:
>
> http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=61841
>
> The redhat engineer suggested that since it was a feature request it was
> to late in the development cycle that it wouldn't make it into RH 7.3, but
> that it could be more easily incorporated if the next release of nfsutils
> if they were already in the upstream source.
>
I checked a slightly different patch. Please check it out and make
sure I didn't break anything.
Thanks.
H.J.
---
2002-04-07 H.J. Lu <[email protected]>
* etc/redhat/nfs.init: Read /etc/sysconfig/nfsd for NFS tuning.
Index: etc/redhat/nfs.init
===================================================================
RCS file: /cvsroot/nfs/nfs-utils/etc/redhat/nfs.init,v
retrieving revision 1.14
diff -u -p -r1.14 nfs.init
--- etc/redhat/nfs.init 25 Jul 2001 16:23:22 -0000 1.14
+++ etc/redhat/nfs.init 7 Apr 2002 20:49:25 -0000
@@ -8,6 +8,7 @@
# networks. This service provides NFS server functionality, \
# which is configured via the /etc/exports file.
# probe: true
+# config: /etc/sysconfig/nfsd
# Source function library.
. /etc/rc.d/init.d/functions
@@ -27,37 +28,76 @@ fi
[ -x /usr/sbin/exportfs ] || exit 0
[ -s /etc/exports ] || exit 0
-RQUOTAD=`type -path rpc.rquotad` # Remote quota server
+# Check for and source configuration file otherwise set defaults
+# TUNE_QUEUE: controls whether to up the size of input queues
+NFSDCFG=/etc/sysconfig/nfsd
+if [ -f "$NFSDCFG" ]; then
+ # Tune nfs server settings
+ . "$NFSDCFG"
+else
+ # Set some defaults
+ TUNE_QUEUE ="no"
+ # Default to NFS version 3.
+ RPCMOUNTDOPTS=""
+fi
+
+# Number of servers to be started by default
+[ -z "$RPCNFSDCOUNT" ] && RPCNFSDCOUNT=8
+
+# Remote quota server
+[ -z "$RQUOTAD" ] && RQUOTAD=`type -path rpc.rquotad`
-# Number of servers to be started up by default
-RPCNFSDCOUNT=8
-# Default to NFS version 3.
-RPCMOUNTDOPTS=""
+# Get the initial values for the input sock queues
+# at the time of running the script.
+if [ "$TUNE_QUEUE" = "yes" ]; then
+ RMEM_DEFAULT=`/sbin/sysctl -n net.core.rmem_default`
+ RMEM_MAX=`/sbin/sysctl -n net.core.rmem_max`
+ WMEM_DEFAULT=`/sbin/sysctl -n net.core.wmem_default`
+ WMEM_MAX=`/sbin/sysctl -n net.core.wmem_max`
+ # 256kb recommended minimum size based on SPECsfs NFS benchmarks
+ [ -z "$NFS_QS" ] && NFS_QS=262144
+fi
# See how we were called.
case "$1" in
start)
# Start daemons.
+ # Apply input queue increase for nfs server
+ if [ "$TUNE_QUEUE" = "yes" ]; then
+ /sbin/sysctl -w net.core.rmem_default=$NFSD_QS >/dev/null 2>&1
+ /sbin/sysctl -w net.core.rmem_max=$NFSD_QS >/dev/null 2>&1
+ /sbin/sysctl -w net.core.wmem_default=$NFSD_QS >/dev/null 2>&1
+ /sbin/sysctl -w net.core.wmem_max=$NFSD_QS >/dev/null 2>&1
+ fi
action "Starting NFS services: " /usr/sbin/exportfs -r
- if [ -n "$RQUOTAD" ]; then
- echo -n "Starting NFS quotas: "
- daemon rpc.rquotad
- echo
+ if [ -n "$RQUOTAD" -a "$RQUOTAD" != "no" ]; then
+ echo -n "Starting NFS quotas: "
+ daemon rpc.rquotad
+ echo
fi
echo -n "Starting NFS daemon: "
daemon rpc.nfsd $RPCNFSDCOUNT
echo
- # Let's see if we support NFS version 3.
- /usr/sbin/rpcinfo -u localhost nfs 3 &>/dev/null
- if [ $? -ne 0 ]; then
+ if [ -z "$RPCMOUNTDOPTS" ]; then
+ # Let's see if we support NFS version 3.
+ /usr/sbin/rpcinfo -u localhost nfs 3 &>/dev/null
+ if [ $? -ne 0 ]; then
RPCMOUNTDOPTS="--no-nfs-version 3"
+ fi
fi
echo -n "Starting NFS mountd: "
daemon rpc.mountd $RPCMOUNTDOPTS
echo
touch /var/lock/subsys/nfs
+ # reset input queue for rest of network services
+ if [ "$TUNE_QUEUE" = "yes" ]; then
+ /sbin/sysctl -w net.core.rmem_default=$RMEM_DEFAULT >/dev/null 2>&1
+ /sbin/sysctl -w net.core.rmem_max=$RMEM_MAX >/dev/null 2>&1
+ /sbin/sysctl -w net.core.wmem_default=$WMEM_DEFAULT >/dev/null 2>&1
+ /sbin/sysctl -w net.core.wmem_max=$WMEM_MAX >/dev/null 2>&1
+ fi
;;
stop)
# Stop daemons.
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
On Sunday April 7, [email protected] wrote:
>
> I checked a slightly different patch. Please check it out and make
> sure I didn't break anything.
...
> + if [ "$TUNE_QUEUE" = "yes" ]; then
> + /sbin/sysctl -w net.core.rmem_default=$NFSD_QS >/dev/null 2>&1
> + /sbin/sysctl -w net.core.rmem_max=$NFSD_QS >/dev/null 2>&1
> + /sbin/sysctl -w net.core.wmem_default=$NFSD_QS >/dev/null 2>&1
> + /sbin/sysctl -w net.core.wmem_max=$NFSD_QS >/dev/null 2>&1
> + fi
Though it possibly doesn't hurt, there is nothing to be gained by
setting the wmem_{default,max}.
Only the rmem needs to be set. This is because nfsd doesn't take
packets off the udp queue until they are completely processed, so the
rmem_{default,max} needs to be big enough to hold all currently active
requests.
NeilBrown
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
On Mon, Apr 08, 2002 at 12:40:55PM +1000, Neil Brown wrote:
> On Sunday April 7, [email protected] wrote:
> >
> > I checked a slightly different patch. Please check it out and make
> > sure I didn't break anything.
>
> ...
> > + if [ "$TUNE_QUEUE" = "yes" ]; then
> > + /sbin/sysctl -w net.core.rmem_default=$NFSD_QS >/dev/null 2>&1
> > + /sbin/sysctl -w net.core.rmem_max=$NFSD_QS >/dev/null 2>&1
> > + /sbin/sysctl -w net.core.wmem_default=$NFSD_QS >/dev/null 2>&1
> > + /sbin/sysctl -w net.core.wmem_max=$NFSD_QS >/dev/null 2>&1
> > + fi
>
> Though it possibly doesn't hurt, there is nothing to be gained by
> setting the wmem_{default,max}.
> Only the rmem needs to be set. This is because nfsd doesn't take
> packets off the udp queue until they are completely processed, so the
> rmem_{default,max} needs to be big enough to hold all currently active
> requests.
>
I guess I don't have to tell you feel free to make any necessary
changes :-).
H.J.
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs