[Top] [All Lists]

Re: error making real-time filesystem + logging question

To: lord@xxxxxxx, roy@xxxxxxxxxxxxx
Subject: Re: error making real-time filesystem + logging question
From: Roy Sigurd Karlsbakk <roy@xxxxxxxxxxxxx>
Date: Tue, 04 Dec 2001 19:50:48 +0000
Cc: rabe@xxxxxxxxxxxxxx, linux-xfs@xxxxxxxxxxx, stimits@xxxxxxxxxx
Reply-to: roy@xxxxxxxxxxxxx
Sender: owner-linux-xfs@xxxxxxxxxxx
>The XFS realtime subvolume is a separate dataspace in the filesystem, it
>is only used for files marked as realtime after they are created. There
>is nothing in the realtime subvolume except file data, where as in the
>normal XFS data subvolume there is a mix of metadata and file data.
>Space allocation in the realtime subvolume is managed by a different
>allocator which gives out space in multiples of a pre-specified size,
>the allocator uses a binary chop approach to space allocation which
>is designed to avoid fragmentation at the expense of efficient use of
>space - the allocator should also be faster.
>The end result is that streaming I/O should behave better on realtime
>files and on regular files.

ok. But XFS realtime subvolumes are, as I understand, not yet supported on 
platform. Are there any plans of supporting it soon?

How is a file placed or retrieved on the realtime subvolume? Is it flagged as 
been created, or does the realtime-part and the data-part come up as separate 
with separate mount points?

>One feature which can be used on normal files which might help with
>placement is preallocation of space. Take a look at the use of the
>XFS_IOC_RESVSP64 ioctl in cmd/xfsprogs/mkfile/xfs_mkfile.c

Is this available through /proc ?


Roy Sigurd Karlsbakk, MCSE, MCNE, CLS, LCA

Computers are like air conditioners.
They stop working when you open Windows.

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