Questions on NFS client inode management.

Questions on NFS client inode management.

[chen, xiangping]

Hi,

I have a question on inode management for NFS client.
It seems that the inodes created on a NFS client for 
a mounted nfs file system stays around until the file 
being removed. Is there any limits on how many inodes
are allowed in memory for NFS? What kind of behavior
we expect if a malicious/careless application just 
keeps creating new files and flood the kernel memory 
with inodes created?

Thanks,

Xiangping 

Re: Questions on NFS client inode management.

[Trond Myklebust]

>>>>> " " == xiangping chen <chen> writes:

     > Hi, I have a question on inode management for NFS client.  It
     > seems that the inodes created on a NFS client for a mounted nfs
     > file system stays around until the file being removed. Is there
     > any limits on how many inodes are allowed in memory for NFS? 
     > What kind of behavior we expect if a malicious/careless
     > application just keeps creating new files and flood the kernel
     > memory with inodes created?

The same behaviour as for local files: as memory goes low, the
ordinary reclaiming schemes kick in, and all should be hunky-dory. Of
course in the *real world*, ...

Cheers,
   Trond

RE: Questions on NFS client inode management.

[chen, xiangping]

Hi, Trond

Thanks for the reply. What kind of reclaiming scheme is used by
NFS client and the Linux local file system, say ext2/ext3? 
Is it used for reclaiming the inode resources, or for memory
in general (like killing processes)?

Is there any benchmark available to test how many active inodes
a particular system can support?


Xiangping


-----Original Message-----
From: Trond Myklebust [mailto:trond.myklebust@fys.uio.no]
Sent: Tuesday, September 11, 2001 4:22 PM
To: chen, xiangping
Cc: 'nfs-request@lists.sourceforge.net'; linux-kernel@vger.kernel.org
Subject: Re: Questions on NFS client inode management.


>>>>> " " == xiangping chen <chen> writes:

     > Hi, I have a question on inode management for NFS client.  It
     > seems that the inodes created on a NFS client for a mounted nfs
     > file system stays around until the file being removed. Is there
     > any limits on how many inodes are allowed in memory for NFS? 
     > What kind of behavior we expect if a malicious/careless
     > application just keeps creating new files and flood the kernel
     > memory with inodes created?

The same behaviour as for local files: as memory goes low, the
ordinary reclaiming schemes kick in, and all should be hunky-dory. Of
course in the *real world*, ...

Cheers,
   Trond

RE: Questions on NFS client inode management.

[Trond Myklebust]

>>>>> " " == xiangping chen <chen> writes:

     > Hi, Trond Thanks for the reply. What kind of reclaiming scheme
     > is used by NFS client and the Linux local file system, say
     > ext2/ext3? Is it used for reclaiming the inode resources, or
     > for memory in general (like killing processes)?

I'm not really much of an expert on the memory management. I try to
stick to doing NFS. If you are interested in the exact mechanisms of
page reclaiming, you might want to try to study the function
try_to_free_pages() in mm/vmscan.c. It never stops changing though ;-)


     > Is there any benchmark available to test how many active inodes
     > a particular system can support?

Not to my knowledge, but as I say, I'm not really an expert...

Cheers,
   Trond