Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 14 Apr 2002 15:24:33 -0400
From:      "Gregory Keefe" <keefeg@keefeg.com>
To:        "Fernando Gleiser" <fgleiser@cactus.fi.uba.ar>
Cc:        <freebsd-questions@freebsd.org>
Subject:   Re: Softupdates
Message-ID:  <00cc01c1e3ea$03382c70$9865fea9@GPC>
References:  <20020414122514.E5464-100000@cactus.fi.uba.ar>

next in thread | previous in thread | raw e-mail | index | archive | help
This is a multi-part message in MIME format.

------=_NextPart_000_00C9_01C1E3C8.7A9F5030
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable


> > FreeBSD Claim:
> > http://www.freebsd.org/features.html
> > Soft Updates allows improved file system performance without =
sacrificing
> > safety and reliability
> >
> > A Unix Expert's Claim:
> > http://cr.yp.to/qmail/faq/reliability.html
> > ``Do not use async or softupdates filesystems. If you do, and if =
your system
> > crashes at the wrong moment, you will lose [data].''
>
> Sofupdates ensure that the *filesystem* is not corrupt after a crash.
> They don't say anything about *data*. You may lose data after a crash.
> If you are worried about losing data, you can mount the filesystems
> sync, but I don't think you'll like the performance. =3D0)

I suspect the world widely acknowledges FreeBSD as a strong server =
platform, at least more so than as a strong workstation platform (I use =
it for one of my workstations, but at work everybody uses Windows).  If =
you accept that premise, then does it follow that the default install of =
FreeBSD should be geared toward a server instead of a workstation?  As a =
server, what are its most common deployment scenarios?  Web servers & =
mail servers?  Under a standard web or mail server setup, is the disk =
I/O the most likely bottleneck?  If so, then I suspect leaving =
softupdates turned on by default is reasonable.  If the bottleneck is =
something else (bandwidth, or perhaps even no bottleneck at all under =
most setups), then does it not make sense to offer the most reliability =
that's possible, at least by default?

Looking at DJB's claim again up top, I suspect that he desires his =
software to really "mean it" when it responds to a client saying it =
successfully received mail.  And the best definition of "mean it" in =
this context is that the mail is safely written to disk (I'd personally =
go a step further and have it safely written to a cluster of servers' =
disks, but that's certainly not reasonable for most environments).  If =
softupdates indeed compromises that feature, though, then couldn't a =
faint shadow of doubt be cast over the reliability of the entire default =
install?

> IMHO, filesystem corruption is far worse than data loss of the last =
writes
> before the crash.

I believe that both are bad.

> Use softupdates on your filesystems and use a good UPS and hardware to
> minimize the risk of a crash.
>=20
> You have a good UPS on your production systems, dont you =3D0)

Yes, I do.  I realize that for most installs this isn't a significant =
issue.  I just wish http://www.freebsd.org/features.html would better =
explain that softupdates doesn't sacrifice filesystem integrity, instead =
of broadly saying softupdates doesn't sacrifice "reliability" which in =
some ways it apparently does.


------=_NextPart_000_00C9_01C1E3C8.7A9F5030
Content-Type: text/html;
	charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META http-equiv=3DContent-Type content=3D"text/html; =
charset=3Diso-8859-1">
<META content=3D"MSHTML 6.00.2715.400" name=3DGENERATOR>
<STYLE></STYLE>
</HEAD>
<BODY>
<DIV><FONT face=3DArial size=3D2></FONT>&nbsp;</DIV>
<DIV><FONT face=3DArial size=3D2>&gt; &gt; FreeBSD Claim:<BR>&gt; &gt; =
</FONT><A=20
href=3D"http://www.freebsd.org/features.html"><FONT face=3DArial=20
size=3D2>http://www.freebsd.org/features.html</FONT></A><BR><FONT =
face=3DArial=20
size=3D2>&gt; &gt; Soft Updates allows improved file system performance =
without=20
sacrificing<BR>&gt; &gt; safety and reliability<BR>&gt; &gt;<BR>&gt; =
&gt; A Unix=20
Expert's Claim:<BR>&gt; &gt; </FONT><A=20
href=3D"http://cr.yp.to/qmail/faq/reliability.html"><FONT face=3DArial=20
size=3D2>http://cr.yp.to/qmail/faq/reliability.html</FONT></A><BR><FONT =
face=3DArial=20
size=3D2>&gt; &gt; ``Do not use async or softupdates filesystems. If you =
do, and=20
if your system<BR>&gt; &gt; crashes at the wrong moment, you will lose=20
[data].''<BR>&gt;</FONT></DIV>
<DIV><FONT face=3DArial size=3D2>&gt; Sofupdates ensure that the =
*filesystem* is not=20
corrupt after a crash.<BR>&gt; They don't say anything about *data*. You =
may=20
lose data after a crash.<BR>&gt; If you are worried about losing data, =
you can=20
mount the filesystems<BR>&gt; sync, but I don't think you'll like the=20
performance. =3D0)<BR></FONT></DIV>
<DIV><FONT face=3DArial size=3D2>I suspect the world widely acknowledges =
FreeBSD as=20
a strong server platform, at least&nbsp;more so than as a=20
strong&nbsp;workstation platform (I use it for one of my workstations, =
but at=20
work everybody uses Windows).&nbsp; If you accept that premise, then =
does it=20
follow that the default install of FreeBSD should be geared toward a =
server=20
instead of a workstation?&nbsp; As a server, what are its most common =
deployment=20
scenarios?&nbsp;&nbsp;Web servers &amp;&nbsp;mail servers?&nbsp; Under a =

standard web or mail server&nbsp;setup, is the disk I/O the most likely=20
bottleneck?&nbsp; If so, then I suspect leaving softupdates turned on by =
default=20
is reasonable.&nbsp; If the bottleneck is something else (bandwidth, or =
perhaps=20
even no bottleneck at all under most setups), then does it not make =
sense to=20
offer the most reliability that's possible, at least by =
default?</FONT></DIV>
<DIV><FONT face=3DArial size=3D2></FONT>&nbsp;</DIV>
<DIV><FONT face=3DArial size=3D2>Looking at DJB's claim again up top, I =
suspect that=20
he desires his software to really "mean it" when it&nbsp;responds to=20
a&nbsp;client&nbsp;saying it successfully received&nbsp;mail.&nbsp; And =
the best=20
definition of "mean it" in this context is that the mail is safely =
written to=20
disk (I'd personally go a step further and have it safely written to a =
cluster=20
of servers' disks, but that's certainly not reasonable for most=20
environments).&nbsp; If softupdates indeed compromises that=20
feature,&nbsp;though, then couldn't a&nbsp;faint shadow of doubt&nbsp;be =
cast=20
over the reliability of the entire default install?</FONT></DIV>
<DIV><FONT face=3DArial size=3D2></FONT>&nbsp;</DIV>
<DIV><FONT face=3DArial size=3D2>&gt; IMHO, filesystem corruption is far =
worse than=20
data loss of the last writes<BR>&gt; before the crash.<BR></FONT></DIV>
<DIV><FONT face=3DArial size=3D2>I believe that&nbsp;both are =
bad.</FONT></DIV>
<DIV><FONT face=3DArial size=3D2></FONT>&nbsp;</DIV>
<DIV><FONT face=3DArial size=3D2>&gt; Use softupdates on your =
filesystems and use a=20
good UPS and hardware to<BR>&gt; minimize the risk of a crash.<BR>&gt; =
<BR>&gt;=20
You have a good UPS on your production systems, dont you =
=3D0)<BR></FONT></DIV>
<DIV><FONT face=3DArial size=3D2>Yes, I do.&nbsp; I realize that for =
most installs=20
this isn't&nbsp;a significant&nbsp;issue.&nbsp; I just wish <A=20
href=3D"http://www.freebsd.org/features.html"><FONT face=3DArial=20
size=3D2>http://www.freebsd.org/features.html</FONT></A>&nbsp;would =
better explain=20
that&nbsp;softupdates doesn't sacrifice filesystem integrity, instead of =
broadly=20
saying softupdates doesn't&nbsp;sacrifice "reliability" which in some =
ways it=20
apparently does.</FONT></DIV>
<DIV><FONT face=3DArial size=3D2></FONT><FONT face=3DArial=20
size=3D2></FONT>&nbsp;</DIV></BODY></HTML>

------=_NextPart_000_00C9_01C1E3C8.7A9F5030--


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




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00cc01c1e3ea$03382c70$9865fea9>