home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pmafire!news.dell.com!swrinde!cs.utexas.edu!hellgate.utah.edu!fcom.cc.utah.edu!osiris.usi.utah.edu!ben
- From: ben@osiris.usi.utah.edu (Ben Pratt)
- Newsgroups: comp.unix.aix
- Subject: JFS i-nodes summary
- Keywords: jfs i-nodes filsystem
- Message-ID: <1992Aug28.174656.8854@fcom.cc.utah.edu>
- Date: 28 Aug 92 17:46:56 GMT
- Sender: news@fcom.cc.utah.edu
- Organization: Utah Supercomputing Institute
- Lines: 58
-
- Several people have asked me to summarize the information I received
- concerning allocation of i-nodes in a journaled filesystem.
- Unfortunately, I've received only one response and it basically a
- re-phrasing of my post... So it looks like there are plenty of folks
- with questions but few with answers.
-
- There has been another post here recently concerning where one might find
- papers or informal documents on the jfs internals, and I too would be
- interested in finding these, if they indeed exist for public consumption.
-
- What I have determined heuristically (pure experience and
- experimentation):
-
- (1) i-nodes in a jfs are statically allocated. I had hoped that
- even thought the i-nodes were preallocated, if you started
- exceeding dataspace and had plenty of i-nodes, the system would
- diminish i-nodes to compenstate...) This was verified by using
- one of my 2 gig filesystems --- I calculated the size of the data
- blocks available by:
-
- X = Space available reported by df
- Y = Space in use reported by 'du -sk /fs'
- Z = Size of filesystem (2Gig) in K.
-
- Theoretical_File_Limit = Z - ( X + Y ) ;
- Test_File_Limit = Z - X ;
-
- Using a C program and dd I could create a file of
- Test_File_Limit. I could not make it any larger, showing that
- I-nodes were not de-allocated dynamically and supporting the idea
- that they must then be allocated statically.
-
- (2) Given that i-nodes are 128 bytes, it became obvious from
- 'doing the arithmetic' that AIX pre-allocates ONE I-NODE FOR
- EVERY 4K BLOCK. In our situation, that means that roughly 65
- Megabytes of a 2Gig filesystem are used up in i-nodes. We figure
- that, given our situation and needs, that is a waste of 55 to 60
- meg per 2 gig filesystem. Since we have 8 of them, that means we
- are losing 440 Megabytes to 480 Megabytes of scratch file space
- for our applications... Ouch.
-
-
- Can any Austin IBM'ers or others verify or refute these claims? Is there
- an alternative method for getting more out our filesystems? Is there
- away around static pre-allocation of i-nodes?
-
- Before the flames fly, let me say that I am not complaining (really). I
- just want an accurate picture of the problem and anyone's input on
- possible solutions. We have a very specific need for large filesystems
- with few files -- hardly a generic situation. Most will find the 'i-node
- per 4k block' acceptable (I assume) in the general case.
-
- Looking forward to any feedback,
-
- Ben Pratt
- Utah Supercomputing Institute
- University of Utah
- ben@snow.usi.utah.edu
-