Date: Tue, 3 Apr 2001 20:59:20 +0200 (CEST) From: bg@sics.se To: FreeBSD-gnats-submit@freebsd.org Cc: bg@sics.se Subject: kern/26324: Slow NFS TCP mounts Message-ID: <200104031859.f33IxKm19862@bg.sics.se>
next in thread | raw e-mail | index | archive | help
>Number: 26324
>Category: kern
>Synopsis: Defaults for NFS mounts over TCP are slow
>Confidential: no
>Severity: non-critical
>Priority: medium
>Responsible: freebsd-bugs
>State: open
>Quarter:
>Keywords:
>Date-Required:
>Class: change-request
>Submitter-Id: current-users
>Arrival-Date: Tue Apr 03 12:00:03 PDT 2001
>Closed-Date:
>Last-Modified:
>Originator: Bjoern Groenvall
>Release: FreeBSD 4.2-RELEASE i386
>Organization:
SICS
>Environment:
FreeBSD NFS clients.
>Description:
Defaults for NFSv3 mounts over TCP limits rsize and wsize to
8kbytes. This is unnecessarily conservative and leads to slow
NFS read/write performance.
NFSv3 has a mechanism to negotiate these parameters so there is no
danger in increasing the defaults to something more appropiate (e.g
sun defaults to 32kbytes).
>How-To-Repeat:
>Fix:
The following patch increases NFS read and write performance between
two FreeBSD machines to ~80Mbytes/s. This was measured over an idle
100Mbit/s network.
Patch was tested on a FreeBSD 4.2-RELEASE machine.
--- /sys/nfs/nfs_vfsops.c.ORIG Sun Sep 10 03:45:36 2000
+++ /sys/nfs/nfs_vfsops.c Mon Apr 2 14:13:35 2001
@@ -841,7 +841,6 @@
register struct nfsmount *nmp;
struct nfsnode *np;
int error;
- struct vattr attrs;
if (mp->mnt_flag & MNT_UPDATE) {
nmp = VFSTONFS(mp);
@@ -883,8 +882,17 @@
nmp->nm_timeo = NFS_TIMEO;
nmp->nm_retry = NFS_RETRANS;
- nmp->nm_wsize = NFS_WSIZE;
- nmp->nm_rsize = NFS_RSIZE;
+ if (argp->flags & NFSMNT_NFSV3) {
+ if (argp->sotype == SOCK_STREAM) {
+ nmp->nm_wsize = nmp->nm_rsize = NFS_MAXDATA;
+ } else {
+ nmp->nm_wsize = NFS_WSIZE;
+ nmp->nm_rsize = NFS_RSIZE;
+ }
+ } else {
+ nmp->nm_wsize = NFS_WSIZE;
+ nmp->nm_rsize = NFS_RSIZE;
+ }
nmp->nm_readdirsize = NFS_READDIRSIZE;
nmp->nm_numgrps = NFS_MAXGRPS;
nmp->nm_readahead = NFS_DEFRAHEAD;
@@ -932,10 +940,15 @@
*vpp = NFSTOV(np);
/*
- * Get file attributes for the mountpoint. This has the side
- * effect of filling in (*vpp)->v_type with the correct value.
+ * Retrieval of mountpoint attributes is delayed until nfs_root
+ * or nfs_statfs are first called. This will happen either when
+ * we first traverse the mount point or if somebody does a df.
+ *
+ * NFSSTA_GOTFSINFO is used to flag if we have succesfully
+ * retreived mountpoint attributes. In the case of NFSv3 we
+ * also flag static fsinfo.
*/
- VOP_GETATTR(*vpp, &attrs, curproc->p_ucred, curproc);
+ (*vpp)->v_type = VNON;
/*
* Lose the lock but keep the ref.
@@ -1042,6 +1055,20 @@
if (error)
return (error);
vp = NFSTOV(np);
+ /*
+ * Get transfer parameters and root vnode attributes.
+ */
+ if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
+ if (nmp->nm_flag & NFSMNT_NFSV3) {
+ nfs_fsinfo(nmp, vp, curproc->p_ucred, curproc);
+ mp->mnt_stat.f_iosize = nfs_iosize(nmp);
+ } else {
+ struct vattr attrs;
+ error = VOP_GETATTR(vp, &attrs, curproc->p_ucred, curproc);
+ if (!error)
+ nmp->nm_state |= NFSSTA_GOTFSINFO;
+ }
+ }
if (vp->v_type == VNON)
vp->v_type = VDIR;
vp->v_flag = VROOT;
>Release-Note:
>Audit-Trail:
>Unformatted:
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-bugs" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200104031859.f33IxKm19862>