*BSD News Article 44626


Return to BSD News archive

Xref: sserve comp.unix.bsd.freebsd.misc:1474 news.software.nntp:13670
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.uwa.edu.au!classic.iinet.com.au!news.uoknor.edu!news.ecn.uoknor.edu!paladin.american.edu!gatech!howland.reston.ans.net!news.moneng.mei.com!brasil.moneng.mei.com!not-for-mail
From: jgreco@brasil.moneng.mei.com (Joe Greco)
Newsgroups: comp.unix.bsd.freebsd.misc,news.software.nntp
Subject: Re: Newspool
Date: 26 May 1995 08:39:30 -0500
Organization: Solaria Public Access UNIX - Milwaukee, WI
Lines: 75
Distribution: inet
Message-ID: <3q4lmi$8nm@brasil.moneng.mei.com>
References: <3q20eu$vk@airhk.air.org> <3q2524$7ot@brasil.moneng.mei.com> <3q2tur$ffb@agate.berkeley.edu>
NNTP-Posting-Host: brasil.moneng.mei.com

In news.software.nntp article <3q2tur$ffb@agate.berkeley.edu>, jkh@violet.berkeley.edu (Jordan K. Hubbard) wrote:
:>just fine.... news.sol.net (FreeBSD 2.0R) has a 7GB news spool, done this
:
:You must not be taking a very aggressive feed.  For news spools, I can
:attest from personal experience that 4096/512 is NOT a very good value
:and tends to run you out of inodes pretty quick..  A much better value
:for a news partition is `-i 1024'.  The 2.0.5 installation allows you to
:specify newfs parameters on a per-filesystem basis.

"not be taking a very aggressive feed"?

news.sol.net takes a *full* 20,000+ group Usenet feed.

spool.mu.edu is one of the most influential news servers around, according
to Brian Reid's statistics.  It does not carry some of the regional
hierarchies, but carries the rest.  While it is not a FreeBSD machine, I
used the same techniques to lay it out - and I notice the same average
inode/space statistics:

Filesystem            kbytes    used   avail capacity  iused   ifree  %iused
/dev/dsk/c0t1d0s0     964174  665383  202375    77%   331867  151461    69%
/dev/dsk/c0t5d0s0    1969089 1295088  477095    73%   220630  767498    22%

(I do not have numbers for news.sol.net handy, but I do check the
percentages every few months, and can testify that the numbers are
"sufficiently similar").

The first is one of the regular Usenet spool disks.  Notice that %iused
tracks capacity fairly closely, although there is a small reserve of excess
inodes.  The second one is the poor little alt.* disk.  Here, the standard
statistics break down, as expected, with the much larger articles that
comprise alt.binaries throwing the statistics significantly.

When I set that disk up, I had a bit of a dilemma:  I knew it would be
better to newfs with default values, but I am hoping to pull alt.binaries
off onto another disk at some point down the road, and formatting with
4096/512 would allow me to make such a change without re-newfs'ing this
filesystem.

However, I really must have missed the boat here.  How does the selection of
4096/512 "run you out of inodes pretty quick", given that the block and
fragment sizes selected have nothing to do with the number of bytes per
inode?

I can see a case being made for staying with 8192/1024 if you're silly
enough to buy a large disk and you don't want to be able to max out your
disk space capacity (and shake the drive to pieces as much... i.e. justify
another disk to your boss).  Or if it's an alt.binaries partition.  But I
don't see anybody buying large disks, all the full feed sites that I know of
invariably split up the spool over multiple disks (just wait until you have
a few hundred nnrpd clients, mid-day)...  it's the only way to get reasonable
filesystem throughput.  Lots of independent heads.  Maybe if you don't have
dozens of nnrpd clients and sixty nntplinks running, it doesn't matter as
much.  But I really don't see 8192/1024 as anything more than throwing away 
some of your expensive disk real estate, unless you're just trying to "hide"
some of your disk space to justify that extra disk (I know somebody who 
did that at one time  ;-)  )

As for the number of bytes per inode, with the exception of alt.binaries,
where the number could actually be tuned *up*, I see no compelling evidence
that -i 1024 buys you anything.  Do you have a df -i that suggests
otherwise?...  I've been advising people to do this for years, and I'd be
interested to see evidence to the contrary.  ;-)

Have a good weekend.

(P.S. Where is 2.0.5?  :-)  I'm buying a nice shiny all-new news.sol.net
from Rod just so I can run 2.0.5 and avoid some of the random crashes I've
had with 2.0R..)

... Joe

-------------------------------------------------------------------------------
Joe Greco - Systems Administrator			      jgreco@ns.sol.net
Solaria Public Access UNIX - Milwaukee, WI			   414/342-4847