[news.software.b] tuning number of inodes for cnews

ramsey@NPIRS.Purdue.EDU (Ed Ramsey) (03/02/91)

I was wondering if on a sparcstation 1+ running sunos 4.1.1 
more inodes are needed (or would be useful) for the news spool
directory than suninstall provides by default on a 100Mb partition. 

I raise this questions because I was told "you must double the
number of inodes in the news partition", but there is no mention of
this topic in the cnews docs that I can find.

Any guidance would be appreciated.

-Ed

Ed Ramsey   ramsey@npirs.purdue.edu     317/494-6616    fax 317/494-0535
There is nothing more important to me than knowing Jesus as He really is.

sean@utoday.com (Sean Fulton) (03/03/91)

In article <1991Mar1.230651.4574@NPIRS.Purdue.EDU> ramsey@NPIRS.Purdue.EDU (Ed Ramsey) writes:
>I was wondering if on a sparcstation 1+ running sunos 4.1.1 
>more inodes are needed (or would be useful) for the news spool
>directory than suninstall provides by default on a 100Mb partition. 
>
>I raise this questions because I was told "you must double the
>number of inodes in the news partition", but there is no mention of
>this topic in the cnews docs that I can find.
>
	We have a 120 M news partition with something like 65000
inodes on it. I got the suggestion from someone on the net and just
set the inodes to the max. I didn't see it in the docs either.
 

-- 
Sean Fulton					sean@utoday.com
UNIX Today!					(516) 562-5430
 /* The opinions expressed above are not those of my employer */

henry@zoo.toronto.edu (Henry Spencer) (03/03/91)

In article <1991Mar1.230651.4574@NPIRS.Purdue.EDU> ramsey@NPIRS.Purdue.EDU (Ed Ramsey) writes:
>I was wondering if on a sparcstation 1+ running sunos 4.1.1 
>more inodes are needed (or would be useful) for the news spool
>directory than suninstall provides by default on a 100Mb partition. 
>
>I raise this questions because I was told "you must double the
>number of inodes in the news partition", but there is no mention of
>this topic in the cnews docs that I can find.

C News predates SunOS 4.x, and since most of our Suns are still running 3.5,
we don't have much real experience with the matter.  Utzoo, which keeps most
news four days, has about 90MB of news using about 40K inodes.
-- 
"But this *is* the simplified version   | Henry Spencer @ U of Toronto Zoology
for the general public."     -S. Harris |  henry@zoo.toronto.edu  utzoo!henry

jbryans@beach.csulb.edu (Jack Bryans) (03/03/91)

The recent time warp posting spates have run our system against the inode
limit.  News' partition inode usage exceeds space usage by 3-5%.  I wonder if
we can get away w/freeing inodes by a brute force rm'ing of cancel msgs from
control.  Something like egrep ...|xargs ...|rm ...  Would expiration object?
Any other gotchya's?  We're running Dec. patched cnews.

Jack

feigin@inf.ethz.ch (Adam Feigin) (03/04/91)

You should be just fine if you are running SunOS 4.1.1, since by
default the installation gives you plenty of inodes. In the past
(SunOS 4.0.3), you couldn't even increase the number of inodes with
the -i parameter to newfs (it would silently ignore it), but you could
by 'lying' to mkfs about the disk geometry.

I just finished upgrading our mail/news gateway to 4.1.1, and I didn't
know that the installation created a large number of inodes; just to
be safe, I increased the number of inodes by feeding a lower number to
the -i option to newfs (the default is 2048). Now, I have this
partition with scads of inodes! Here's what I had before (default) and
after:

using default (what SunOS installation created for me):

380923 KB	193024 inodes

using -i option to increase # of inodes (I used  -i 1024 when I ran newfs):

356795 KB	386048 inodes

In its current state:

Filesystem            kbytes    used   avail capacity  Mounted on
/dev/xd0e             356795  318295   20660    94%    /var/spool/news

Filesystem             iused   ifree  %iused  Mounted on
/dev/xd0e             118730  267318    31%   /var/spool/news

I dont know if 4.1 created partitions with large numbers of inodes,
but 4.1.1 definitely does. Depending on how much news you keep around,
you will probably be fine with the default partition that the
installation creates for you. Our default expiry is 14 days, but some
groups are expired sooner. I will probably 'recreate' the default
partition when I get some time, since 31% inode usage on a 94% used
disk is a little outlandish, dont you think ?

							/AWF
------------------------------------------------------------------------------
Internet: feigin@inf.ethz.ch				 Adam W. Feigin
UUCP:{backbones}!ethz!feigin 			Manager, Computing Facilities
Mail: Dept. Informatik				Department of Computer Science
      ETH-Zentrum			  Swiss Federal Institute of Technology
      CH-8092 Zurich				       Zurich, Switzerland
      Switzerland
Phone: +41 1 254 7205
      
		"He who knows, speaks not; he who speaks, knows not"

henry@zoo.toronto.edu (Henry Spencer) (03/04/91)

In article <JBRYANS.91Mar3110745@beach.csulb.edu> jbryans@beach.csulb.edu (Jack Bryans) writes:
>...I wonder if
>we can get away w/freeing inodes by a brute force rm'ing of cancel msgs from
>control.  Something like egrep ...|xargs ...|rm ...  Would expiration object?

In general, the contents of control are pretty uninteresting, so you might
want to skip the egrep.  That's one group that ought to have a real short
expiry time, in case you're not doing that already.  As for simply removing
things, no problem:  history entry present but article missing is the normal
state of affairs after a cancellation -- cancel removes the article but it's
way too much trouble to delete the history entry -- so expire considers this
a normal case, and won't even complain.
-- 
"But this *is* the simplified version   | Henry Spencer @ U of Toronto Zoology
for the general public."     -S. Harris |  henry@zoo.toronto.edu  utzoo!henry

rdc30@nmrdc1.nmrdc.nnmc.navy.mil (LCDR Michael E. Dobson) (03/05/91)

In article <1991Mar1.230651.4574@NPIRS.Purdue.EDU> ramsey@NPIRS.Purdue.EDU (Ed Ramsey) writes:
>I was wondering if on a sparcstation 1+ running sunos 4.1.1 
>more inodes are needed (or would be useful) for the news spool
>directory than suninstall provides by default on a 100Mb partition. 
>
>I raise this questions because I was told "you must double the
>number of inodes in the news partition", but there is no mention of
>this topic in the cnews docs that I can find.
>
>Any guidance would be appreciated.
>
We run with 65536 inodes on a 150Meg partition, the max allowed for a single
partition by SysVR3.2 and sometimes that isn't enough, even with an agressive
expire time :-(

-- 
Mike Dobson, Sys Admin for      | Internet: rdc30@nmrdc1.nmrdc.nnmc.navy.mil
nmrdc1.nmrdc.nnmc.navy.mil      | UUCP:   ...uunet!mimsy!nmrdc1!rdc30
AT&T 3B2/600G Sys V R 3.2.2     | BITNET:   dobson@usuhsb or nrd0mxd@vmnmdsc
WIN/TCP for 3B2                 | MCI-Mail: 377-2719 or 0003772719@mcimail.com

jerry@olivey.ATC.Olivetti.Com (Jerry Aguirre) (03/08/91)

In article <1991Mar3.001647.18652@zoo.toronto.edu> henry@zoo.toronto.edu (Henry Spencer) writes:
>we don't have much real experience with the matter.  Utzoo, which keeps most
>news four days, has about 90MB of news using about 40K inodes.

I run 398MB and 141K inodes, about the same ratio.  In general 1
inode per 2K bytes should be about right.

The problem is that sometimes you don't get as many inodes allocated as
you expect, even less than the default value.  There are a number of
limitations.  SYSV file systems limit one to 65K inodes max per file
system.  4.[23]BSD systems have an internal hidden limit on the number
of inodes per cylinder group.  If you have a drive with lots of sectors
per cylinder (most modern winchester drives) then the resulting number
of inodes will be much less than what is advertized as the default.  And
mkfs doesn't inform you of this little detail.

One can sometimes work around the 4.2BSD limitation by using -C 8 to
reduce the size of the cylinder group.  Some system don't like this and
refuse to cooperate.  It is possible to lie about the geometry in
/etc/disktab and achieve the same effect.  Drives with odd numbers of
surfaces and sectors can be particularly troublesome.

The important thing is to verify the ratio after you make the file
system.  Don't just trust that it did what you asked.

huntting@colorado.edu (03/10/91)

jerry@olivey.ATC.Olivetti.Com (Jerry Aguirre) writes:

>I run 398MB and 141K inodes, about the same ratio.  In general 1
>inode per 2K bytes should be about right.

>The problem is that sometimes you don't get as many inodes allocated as
>you expect, even less than the default value.

Yeow!  Can someone enlighten me on how you check the number of inodes
and data blocks under ultrix 4.1?

I'm bringing up a DS3100 with a wrenV (one spool partition)...


brad
	huntting@colorado.edu
	huntting@csn.org

urlichs@smurf.sub.org (Matthias Urlichs) (03/11/91)

In news.software.b, article <1991Mar10.035342.14226@csn.org>,
  huntting@colorado.edu writes:
< 
< Yeow!  Can someone enlighten me on how you check the number of inodes
< and data blocks under ultrix 4.1?
< 
Mount the partition and "df -i" it.

You have made a mistake whenever the percantage of free inodes is smaller
than the percentage of free space.

NB: If you're using a partition solely for /usr/spool/news, you might use
tunefs to decrease the amount of space held for root. The default is 10%, and
the manual suggests that a reasonable value would be 3% beyond which
performance starts to degrade.
-- 
Matthias Urlichs -- urlichs@smurf.sub.org -- urlichs@smurf.ira.uka.de     /(o\
Humboldtstrasse 7 - 7500 Karlsruhe 1 - FRG -- +49-721-621127(0700-2330)   \o)/

bdale@col.hp.com (Bdale Garbee) (03/14/91)

>NB: If you're using a partition solely for /usr/spool/news, you might use
>tunefs to decrease the amount of space held for root. The default is 10%, and
>the manual suggests that a reasonable value would be 3% beyond which
>performance starts to degrade.

We routinely use 5% minfree and inode/8k for routine filesystems under HP-UX
(which is very BSD-ish filesystem wise).  For our C News filesystem, we are
using 5% minfree with inode/2k, and that seems about right.

Bdale