[rrd-developers] Use of madvise / msync kills performance for me

[Bernhard Fischer]

On Fri, Jun 27, 2008 at 01:41:59PM -0700, Marcus Reid wrote:
>On Fri, Jun 27, 2008 at 01:13:17PM +0200, Bernhard Fischer wrote:
>> On Fri, Jun 27, 2008 at 02:58:15AM -0700, Marcus Reid wrote:
>> >
>> >Matt Dillon provided some good information on this subject that I would
>> >like to pass on.  First, a little background..  I discovered that the
>> >long slow msync() calls only happen on files over a certain size.  The
>> >file that's slowing things down is 1161mB long, and msync() calls to a
>> >file that's 940mB long are fast.  That's probably a kernel problem that
>> >could be worth looking into.
>> 
>> eh, that's quite big, an order of magnitudes bigger than my files, at
>> least.
>> 
>> >
>> >This is probably an edge case -- I'm updating an rrd file that's over
>> >a gig in size and I don't know how common that is.
>
>Well, this is starting to feel like beating on a dead horse, but
>I've been doing some more testing and msync() makes things really
>slow for smaller files as well.
>
>I think the main difference between the platforms is that under
>FreeBSD, msync() even with MS_ASYNC doesn't return until the data
>is on the physical disk, whereas with Linux it's just scheduled to
>make it there at some point.  I'm not sure that this behavior will
>change due to the OpenGroup spec that Matt mentioned, but I will
>bring the issue up again.

SUS requires that an MS_ASYNC and MS_SYNC is available. If your ASYNC
impl is broken (i.e. behaves as if it was SYNC) then you should really
fix your VM :)

Of course we (userspace) can avoid calling msync on __FreeBSD__, but
that's not exactly the proper thing to do, imo.

>I'm just going to file a PR and try and get the msync() call
>patched out in the ports collection.

Sure, but that's wrong. See above.