Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 10 Feb 2002 01:31:39 -0500
From:      Brian T.Schellenberger <bts@babbleon.org>
To:        Sam <sam@wa4phy.net>, freebsd-questions@FreeBSD.ORG
Cc:        freebsd-stable@FreeBSD.ORG
Subject:   Re: Normal behavior?
Message-ID:  <20020210063140.9733B408E@i8k.babbleon.org>
In-Reply-To: <3C5D4702.661789BA@vortex.wa4phy.net>
References:  <3C5D4702.661789BA@vortex.wa4phy.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Sunday 03 February 2002 09:19 am, Sam wrote:
> I'd appreciate some clarification about the behavior of the softupdates
> as it relates to disk writes.  Situation:  Have noticed that during an
> ftp session with relatively high (150 Kbs) data stream, I notice after
> what appears to be a short burst, i.e., 130 packets (viewed with systat
> -vm) there is a considerable delay before the next batch.  Graphicaly
> viewing thruput with xsysinfo, I watch the disk write for each "batch",
> but while the write is happening, there appears to be a significant
> delay before the next packet stream is graphicaly displayed.  Is this a
> function of softupdates, combined with the fact that the ATA drive has
> to be serviced by the processor, and the switcher can't service two
> things at once, or what.  How often does the data that needs to be
> written actually get written to the disk, or do I misunderstand how
> softupdates works?  Essentially, what it appears to me that is
> happening, is the packet stream is "suspended" while the disk is being
> written to.  Is that a correct assumption?  Since softupdates is on by
> default now, what damage would I do if I turned it off?  Is that to my
> best interest?

softupdates will "batch up" writes to the disk that otherwise would be done 
one at a time.  So the total time taken up by disk writes shouldn't be any 
larger, and if you turn off write caching it should actually be lots smaller 
by dint of the batching up.  If you have write caching turned on (and I do 
not recommend this practice with typical IDE disks!), then it should be the 
same speed either way, just batched up vs. spread out.

When interspersed with network i/o, it is possible that your total throughput 
would be faster if you turn them, if the timing is such that the disk i/o 
happens when the network i/o is waiting anyway, and vice-versa, but in 
practice it seems unlikely that the timing will always work out quite so 
conveniently.

Most likely it would be the same or slower if you turn off softupdates, as 
you get just as much "drag" from the disk i/o, but in a less "bursty" fashion.

But the only way to know is to do repeated before-and-after testing.  If you 
do, please post the results back here.  It should prove interesting either 
way.


>
> Thanks..
>
> Sam

-- 
Brian T. Schellenberger . . . . . . .   bts@wnt.sas.com (work)
Brian, the man from Babble-On . . . .   bts@babbleon.org (personal)
                                ME -->  http://www.babbleon.org
http://www.eff.org   <-- GOOD GUYS -->  http://www.programming-freedom.org 

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20020210063140.9733B408E>