Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!purdue!bu-cs!buengc!bph
From: bph@buengc.BU.EDU (Blair P. Houghton)
Newsgroups: comp.unix.questions
Subject: Re: unix question: files per directory
Message-ID: <2531@buengc.BU.EDU>
Date: 15 Apr 89 10:19:01 GMT
References: <24110@beta.lanl.gov> <16839@mimsy.UUCP> <9195@alice.UUCP> <127@dg.dg.com>
Reply-To: bph@buengc.bu.edu (Blair P. Houghton)
Followup-To: comp.unix.questions
Distribution: na
Organization: Boston Univ. Col. of Eng.
Lines: 44

In article <127@dg.dg.com> rec@dg.UUCP (Robert Cousins) writes:
>
>This brings up one of the major physical limiations of the System V 
>file system:  if you can have 2 ^ 24 blocks, and only 2 ^ 16 discrete
>files, then to harness the entire file system space, each file will
>(on average) have to be 2 ^ 8 blocks long or 128 K.  Since we know that
>about 85% of all files on most unix systems are less than 8K and about 
>half are under 1K, I personnally feel that the 16 bit inode number is
>a severe handicap.
>
>Robert Cousins
>
>Speaking for myself alone.

I'll stand behind you, big guy.  I just hacked up a program to check out my
filesizes, and I'll be damned if I didn't think my thing was real big...

On the system I checked (the only one where I'm remotely "typical" :),
I have 854 files, probably two dozen of them zero-length (the result of
some automated VLSI-data-file processing).  The mean is 10.2k, stdev is 60k
(warped by a few megabyte-monsters), and the median is 992 bytes (do
you also guess peoples' weight? :)  Of these 854 files of mine, 84% are
under 8000 bytes, and a paltry eight exceed the 128k "manufacturer's
suggested inode load" you compute above.

For another machine:
1740 files
Median 1304 bytes
Mean   7752
StDev 28857
77% < 8kB
And only 4 (that's FOUR) over the 128k optimal mean.

Hrmph.  And I thought I was more malevolent than that.  At least the
sysadmins can't accuse me of being a rogue drain on the resources...

Consider that "block" can be 1,2,4kB or more, and you're talking some
BIIIG files we have to generate to be efficient with those blocknumbers.

				--Blair
				  "...gon' go lick my wounded ego...
				   and ponder ways to make file
				   systems more efficient, or at least
				   more crowded.  ;-)"