*BSD News Article 45316


Return to BSD News archive

Newsgroups: comp.unix.bsd.freebsd.misc
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!simtel!news.kei.com!news.mathworks.com!uunet!in1.uu.net!monoli!csoft!newsserv
From: Igor Sviridov <sia@lot.cs.kiev.ua>
Subject: Re: NFS write performance
X-Return-Path: lotua!lot.cs.kiev.ua!sia
Reply-To: sia-@UA.NET
Organization: Communication Systems Research Team, Kiev, Ukraine
Date: Fri, 9 Jun 1995 21:46:43 GMT
Message-ID: <802730803PCTIN@lot.cs.kiev.ua>
Lines: 43
Sender: news-service@monolit.kiev.ua

In article <3r3tql$e4e@si-nic.hrz.uni-siegen.de> peter@netz.hrz.uni-siegen.de (Peter Merz) writes:

> a few weeks ago I installed SNAP-950412 on my two systems
> (a pentium and a 486). To my surprise I found that NFS
> performance has significantly increased from 1.1.
> But there is a big gap between writing and reading, 
> on my machines
> 	170 KByte/s for writing
> and	800 KByte/s for reading.
> 
> Are there any changes in the new version (2.0.5) concering
> nfs?

If you want to enter dark land of asynchronous nfs, apply following
patch (relative to 2.0R, but trivial to add to any SNAP) and
recompile kernel with ASYNC_NFS option.

I wonder, if this will be sysctl-setable in 2.0.5?

*** nfs_serv.c.orig	Sun Oct  2 19:26:58 1994
--- nfs_serv.c	Fri Jun  9 23:37:03 1995
*************** nfsrv_write(nfsd, mrep, md, dpos, cred, 
*** 575,581 ****
--- 575,585 ----
  	register long t1;
  	caddr_t bpos;
  	int error = 0, rdonly, cache, siz, len, xfer;
+ #ifdef	ASYNC_NFS
+ 	int ioflags = IO_NODELOCKED;
+ #else
  	int ioflags = IO_SYNC | IO_NODELOCKED;
+ #endif
  	char *cp2;
  	struct mbuf *mb, *mb2, *mreq;
  	struct vnode *vp;

> --
> -- Peter Merz,                                    pmerz@hrz.uni-siegen.de  --
> 

--
Igor