21 messages in com.xensource.lists.xen-develRE: [Xen-devel] [PATCH][TOOLS] Reduci...
FromSent OnAttachments
Graham, Simon12 Feb 2007 13:47.patch
John Levon12 Feb 2007 14:39 
Iustin Pop12 Feb 2007 15:41 
Graham, Simon13 Feb 2007 04:59 
Graham, Simon13 Feb 2007 05:01 
Alan13 Feb 2007 10:49 
Graham, Simon13 Feb 2007 14:12.patch
Keir Fraser21 Feb 2007 02:33 
Graham, Simon21 Feb 2007 07:06 
Keir Fraser21 Feb 2007 07:09 
Graham, Simon21 Feb 2007 07:21 
Robert Read21 Feb 2007 18:10 
Graham, Simon23 Feb 2007 07:13.patch
Keir Fraser23 Feb 2007 08:28 
Graham, Simon23 Feb 2007 10:23 
Graham, Simon23 Feb 2007 10:31 
Keir Fraser23 Feb 2007 10:43 
Graham, Simon23 Feb 2007 10:48 
Keir Fraser23 Feb 2007 15:03 
Graham, Simon23 Feb 2007 16:50 
Graham, Simon24 Feb 2007 06:05.patch
Subject:RE: [Xen-devel] [PATCH][TOOLS] Reducing impactofdomainsave/restore/dump on Dom0
From:Graham, Simon (Simo@stratus.com)
Date:02/23/2007 07:13:33 AM
List:com.xensource.lists.xen-devel
Attachments:

Take 3 -- I've moved the flushing/fadvise-ing code into a common routine -- the only way I found to do this in line with other utilities was to make it inline in xc_private.h (since this is the only file included in both libxenctrl and libxenguest). The 'ifdef __linux__' stuff is now confined to this routine which is better...

I've also left the fsync() in place -- I think it is necessary (and certainly does no harm).

Simon

----------------------------------- Reduce impact of saving/restoring/dumping large domains on Dom0 memory usage by means of fadvise64() to tell the OS to discard the cache pages used for the save/dump file.

-----Original Message----- From: xen-@lists.xensource.com [mailto:xen-devel- boun@lists.xensource.com] On Behalf Of Keir Fraser Sent: Wednesday, February 21, 2007 10:10 AM To: Graham, Simon; Keir Fraser; xen-@lists.xensource.com Subject: Re: [Xen-devel] [PATCH][TOOLS] Reducing impactofdomainsave/restore/dump on Dom0

On 21/2/07 15:06, "Graham, Simon" <Simo@stratus.com> wrote:

2. sync-then-advise is only done at the end of writing a file to ensure that all of the cached pages are discarded. Whilst writing the file, I only fadvise which triggers a write back and discards any clean pages up to the specified offset. This is indeed a performance thing -- fsyncing on every write makes it very slow.

Do you need the fsync at all? It's possible that the kernel will launder-then-discard the affected pages automatically, just from the fadvise() alone.

-- Keir