*BSD News Article 60154


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.mira.net.au!yarrina.connect.com.au!news.mel.connect.com.au!munnari.OZ.AU!news.hawaii.edu!ames!agate!howland.reston.ans.net!newsfeed.internetmci.com!EU.net!Germany.EU.net!Bonn.Germany.EU.net!tools!usenet
From: jk@tools.de (Juergen Keil)
Newsgroups: comp.unix.bsd.netbsd.misc,comp.unix.bsd.bsdi.misc,comp.unix.solaris,comp.unix.aix
Subject: Re: ISP hardware/software choices (performance comparison)
Date: 20 Jan 1996 16:51:20 GMT
Organization: TooLs GmbH, Bonn, Germany
Lines: 59
Distribution: inet
Message-ID: <JK.96Jan20175120@leo.tools.de>
References: <4cmopu$d35@vixen.cso.uiuc.edu> <4d43bt$es8@park.uvsc.edu>
	<4d5vhg$38p@mail.fwi.uva.nl> <4dbun0$j2f@park.uvsc.edu>
	<4de3ml$naq@engnews2.Eng.Sun.COM> <4dgpti$rnv@park.uvsc.edu>
	<4dif2a$6ea@mail.fwi.uva.nl> <4dp5b9$8v8@park.uvsc.edu>
NNTP-Posting-Host: leo.tools.de
In-reply-to: Terry Lambert's message of 19 Jan 1996 22:18:17 GMT
Xref: euryale.cc.adfa.oz.au comp.unix.bsd.netbsd.misc:2071 comp.unix.bsd.bsdi.misc:2227 comp.unix.solaris:57964 comp.unix.aix:69285

In article <4dp5b9$8v8@park.uvsc.edu> Terry Lambert <terry@lambert.org> writes:

> casper@fwi.uva.nl (Casper H.S. Dik) wrote:
> ] How did you disable it and how did you check that it is disabled
> ] on the machine you have access to?
> 
> I disabled it by changing a "yes" to a "no" in the /etc/rc file.

Well, Solaris 2.x doesn't have an /etc/rc file  8-)

And a 
	grep -i yes /etc/{init,rc}.d/* /etc/rc*

reveals about three usages of the string 'yes', but none of them have
anything to do with NFS.  And none of these places where a 'yes'
occurs in these files contains user setable options!

> This is apparently the flag that enables the routine that
> is responsible for old/new NFS write performance; I may be
> able to dig up the information this Saturday via a long
> distance phone call if I have to.

No it's not. The 'flag' I was talking about is a global variable in
the nfs filesystem kernel module (`rfs_write_cluster'), that can be
set to 0 either via /etc/system or an adb patch.  As far as I know,
this flag/variable is completely *undocumented* by sun - it's not in
the solaris 2.3 manual pages and I've found nothing about it in the
Solaris 2.4 answerbooks (sysadmin, sdk, ddk). I don't have access to
Solaris 2.3 answerbooks to check.

I've found about this flag only because I was curious and looked at
the disassembled code for the nfs server write routine after I had
noticed that a SS2 running Solaris 2.3 would write files over NFS much
faster than the same machine running 4.1.x.

> Thanks to Juergen Keil for sending me a private email message
> with the real async write patch.  It is not the patch to which
> I was referring, which I incorrectly identified as the "async
> write patch".

I've send this patch only to show you what dirty hacks you have to
perform on a Sun system (both Solaris 2.x and SunOS 4.1.x) to *enable*
async nfs.

And to emphasize it here: This async nfs patch doesn't come from sun
(I hacked it together on thursday), is evil, is of course completely
unsupported.  It just shows the places in the nfs server write code
that perform the *synchroneous* vnode writes or vnode sync operations.

I don't know what you have 'disabled' on your solaris 2.3 system but
it seems it has nothing to do with 'nfs server caching' (or async nfs
writing) because it was never 'enabled' on your machine.

[Btw: with this patch installed an ELC (which uses the le interface) is
able to write 1080KBytes/sec, so this shows that the le interface can
easily receive data with more than 8 MBit/sec.  As a test I've written
a file that is 10x larger than available memory on the ELC]
-- 
Juergen Keil          jk@tools.de ...!{uunet,mcsun}!unido!tools!jk