xfs
[Top] [All Lists]

Re: inode allocation and volume groups

To: PAulN <pauln@xxxxxxx>
Subject: Re: inode allocation and volume groups
From: Steve Lord <lord@xxxxxxx>
Date: Tue, 17 Feb 2004 06:04:45 -0600
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <40315970.4060801@psc.edu>
References: <40315970.4060801@psc.edu>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mozilla Thunderbird 0.5 (X11/20040208)
PAulN wrote:
Hi,
I know this the linux list but I have an irix question..
I was wondering if someone could please explain the policy for inode allocation
in an XVM environment. To maximize performance, I'd like to create a scheme where
multiple file creates in the same directory are distributed across volume groups. If necessary,
I can resort to creating the inodes in separate directories and hardlinking them into the "proper"
namespace but it would be helpful to know what XFS is doing under the hood.


I've tried various tests to charaterize the behavior but still have yet to fully understand
how it works. Could someone enlighten me?


Thanks,
Paul


Normal behavior is that an inode is placed in the same allocation group as the parent directory. New directories round robin to the next allocation group.

Exceptions to this are:

  o when there is locking contention on the allocation group, or it
    does not have reasonable space available. In this case the allocator
    will move on to the next available allocation group.

  o for filesystems of more than 1 Tbyte inodes can be kept in
    allocation groups within the first 1 Tbyte to keep inode
    numbers down within 32 bits. This can be turned off with
    the inode64 option (except on 32 bit linux where inode
    numbers larger than 32 bits are a bad thing).

Steve


<Prev in Thread] Current Thread [Next in Thread>