From: Jeremy McNicoll <jeremy@mcnicoll.ca>
Subject: Re: RE: [PATCH] 1/3 - RPC metrics support
Date: Wed, 31 Mar 2004 13:25:27 -0500
Sender: nfs-admin@lists.sourceforge.net
Message-ID: <1080757526.3824.8.camel@jrmIBMA30.mcnicoll.ca>
References: <482A3FA0050D21419C269D13989C61130435DE34@lavender-fe.eng.netapp.com>
	 <1080753549.4194.60.camel@lade.trondhjem.org>
Reply-To: jeremy@mcnicoll.ca
Mime-Version: 1.0
Content-Type: text/plain
Cc: Charles Lever <Charles.Lever@netapp.com>, Patrick Mochel <mochel@digitalimplant.org>, nfs@lists.sourceforge.net
To: Trond Myklebust <trond.myklebust@fys.uio.no>
In-Reply-To: <1080753549.4194.60.camel@lade.trondhjem.org>
Errors-To: nfs-admin@lists.sourceforge.net

On Wed, 2004-03-31 at 12:19, Trond Myklebust wrote:
> On Wed, 2004-03-31 at 11:57, Lever, Charles wrote:
> > > My experience with gettimeofday() was that it slows down and 
> > > eventually hangs the machine when you use it inside bottom 
> > > halves. That's why I had to revert to using 'jiffies' 
> > > instead. While I agree that precision may be important, a 
> > > measurement that so impacts on what it is measuring is not 
> > > particularly useful either...
> > 
> > using jiffies makes these measurements useless for anything
> > faster than a millisecond (or 10 milliseconds on current
> > x86), which is most everything on 100Mb or faster, these
> > days.  isn't there some other solution?
> 
> In 2.6. jiffies have 1 ms resolution, not 10 on x86. There may be some
> alternative generic timing solution in Linux that does better than that,
> and that can work within bottom halves without side-effects but I'm not
> aware of such a solution.
> 
> You can of course use do_gettimeofday() for the non bh-safe parts if you
> want to, but using it inside things like *data_ready() etc. was clearly
> unacceptable.
> 
> The only generic solution I see is therefore to use statistics...
> 

I am the middle of the "fun" part of my school project that involves
this stuff, documentation that is!!!  But If I recall correctly there is
a patch that gives you better resolution jiffies.... I cant recall off
the top of my head.  But I have to agree with the jiffies solution.  It
worked really well for me.  It is just a matter of converting jiffies to
MS in user space.  

I am using Robert Love's book right now, for a reference while I
describe what I accomplished. [ (http://www.mcnicoll.ca/iostat/) as soon
as the documentation is completed I will post it there.]   Which has
good info on timing in chapter 9 , this may help you out.   


-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs