Return to BSD News archive
#! rnews 5035 bsd
Path: euryale.cc.adfa.oz.au!platinum.sge.net!como.dpie.gov.au!news.gan.net.au!act.news.telstra.net!vic.news.telstra.net!news.mira.net.au!news.netspace.net.au!news.mel.connect.com.au!news.syd.connect.com.au!news.bri.connect.com.au!corolla.OntheNet.com.au!not-for-mail
From: Tony Griffiths <tonyg@OntheNet.com.au>
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: Out of INODES (some further explanation)
Date: Sat, 28 Jun 1997 11:58:45 +1000
Organization: On the Net (ISP on the Gold Coast, Australia)
Lines: 92
Message-ID: <33B46FD5.5308@OntheNet.com.au>
References: <01bc830e$a0ffc9f0$0f00a8c0@kahuna>
Reply-To: tonyg@OntheNet.com.au
NNTP-Posting-Host: swanee.nt.com.au
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Mailer: Mozilla 3.0 (WinNT; I)
To: Kurt Schafer <kschafer@cyberbeach.net>
Xref: euryale.cc.adfa.oz.au comp.unix.bsd.freebsd.misc:43663
Kurt Schafer wrote:
>
> I received an error that one of the directories I've got set up as a
> repository for USENET articles is out of inodes.
>
> The drive is a 4gig barracuda of which 1.2gigs are taken up. But god knows
> how many actual files and directories are hogging inodes.
>
> The machine is an INND server with the /var/news distributed amongst 4 4gig
> drives via symlinks to different tree hierarchies. (alt rec, etc)
>
> Any pointers would be welcome.
A bit more explanation of your problem (since I have been there, done
that [several times!!!]) ...
Basically, each file in a filesystem requires ONE (1) inode.
A brief decription of an INODE:
Each inode contains the metadata for the file (name, creation date,
access date, owner & protection codes, etc.) as well a pointers (12) to
the data blocks that the file occupies on disk. If the file is 'large'
and occupies more than 12 data blocks, then a form of tree-structured
chaining occurs with top-level pointers pointing to next-level pointer,
etc. until you get to the actual data blocks themselves. If, for
instance, the block size of the filesystem is 8KB (typical /usr
setting), then a file of 96KB or less will have only direct pointers in
the inode!
So, we now come down to filesystem parameter settings (newfs) -
-b block size in bytes (power of 2 from 512 -> ???)
-f fragment size in bytes (1/1, 1/2, 1/4, 1/8 of -b value)
-i inode density in bytes
As stated before, the block is the largest amount of disk that each
pointer can reference. Bigger blocks mean that fewer pointers are
necessary. The 'fragment' is the "unit of allocation" for a filesystem
and is some fraction of a block (1/8th being the smallest fraction).
There is a single bit in the allocation bitmap for each frag in the
filesystem. A fragment is that bit of a file at the end which doesn't
occupy a complete block (eg. a 8193 byte file on a 8KB block'd filesys
will occupy 1 whole block + 1 frag for the 1 extra byte). On average,
1/2 a frag is wasted per file.
The inode density (-i value to newfs) determines how many inodes (128
bytes each) to create for the filesystem. If you choose a value of 4KB
(1 inode per 4096 bytes of filesystem), then a 4GB filesystem will have
1,000,000 (or there abouts) inodes preallocated, and since each takes
128 bytes, roughly 128MB of the 4GB filesystem will be taken up by
inodes. There is other 'overhead' as well, such as the storage bitmap
for frag allocation but inodes represent the largest portion of the
overhead!
Remember that you need 1 inode per file so, depending on the AVERAGE
file size, and number of expected files, you can calculate the number of
inodes (+ some for comfort and peace of mind) you need for a particular
filesystem.
This is what I use for news allocated over 2 x 4GB disks...
/var/spool/news -b 8192 -f 2048 -i 4096
/var/spool/news/alt -b 16384 -f 4096 -i 8192
On the basis of fewer+bigger file on the ".../alt" filesystem, I have
made the blocks and frags bigger and allocated fewer inodes per bytes of
filesystem. Remember that the frag is the smallest unit of allocation
so, on average, 1/2 a frag is wasted per file created. Smaller frags
mean less wasted space but bigger bitmaps to keep track of allocated
space.
There are other parameters to newfs such as "group size" and the like,
but I generally let these default (32MB/group seems to be a favorite
default). The metadata for each group is located at the front of the
group.
For gory details of any existing filesystem, use the command-
# sync; sync; dumpfs /dev/rsd0e > /tmp/usr-fs.info
^^^^^ ^^^^^^^^^^^^^^^^
Your preferences go here!
This will provide a summary of the filesystem followed by info related
to each cylinder-group (blocks, directories, inodes, and frags).
As for a solution to your "out of inodes" problem, as someone else said
it's a matter of trashing the fs and newfs'ing with hopefully better
parameters this time round. With news, it's relatively easy to get it
wrong (several times in my case) and painful to fix each time you do!
Tony