Speed tests

[Paul Haas]

On Thu, 14 Nov 2002, Donovan Baarda wrote:

> On Wed, Nov 13, 2002 at 01:27:46PM +0200, Mozzi wrote:
> > Hi all
> >
> > I hope no one minds but I was asked to post my timetrial findings back
> > to the list.
> > Hope it helps someone else as well, if you have any suggestions please
> > mention them as I need all the speed that I can get
> >
> > The scenaro is as follows
> > I have to switch mail servers and I need to copy all my mbox files over
> > to the new machine.
> > As you may well know time if off the utmost importance so I need to do
> > things as quickly as possible, we are an ISP so we cannot have people
> > offline for to long.
> > I have from 12:00 to 5:00 in the morning
> > I conducted some tests to see what will be the most effecient to copy
> > the data over with and tried to simulate the enviroment as closely as
> > possible but on a smaller scale.
> > Here are my results
> [...]

> In all the tests you are doing a "full copy", so all you are excercising is
> the various methods directory walking and data transfer.

> I'd be interested to see if actualy using the rsync delta for this would
> improve things. Try the following;
>
> 1) rsync the data over from the "live system"

Repeat step 1 and time it.  This will  let you know about how long the
final rsync will take.  It will also make the final rsync faster.

> 2) start the stopwatch... now beginneth the downtime
> 3) rsync the data over again
> 4) stop the stopwatch... now endeth the down time :-)

> Idea being you minimize the transfer by getting a rough copy over first
> without bringing the live system down, then just "update" it during the
> downtime.

> I suspect with gigabit ethernet the "sacrifice CPU for bandwidth" philosopy
> of rsync will mean this is actually worse than the raw copies, but certainly
> for anything slower it should make an improvement.

The timestamp comparison of rsync will be the big win.  If the time stamp
hasn't changed, rsync doesn't even need to open the file.

If reading a file is slower than sending it, full reads and
partial sends don't save much time, if any.

> --
> ----------------------------------------------------------------------
> ABO: finger abo at minkirri.apana.org.au for more info, including pgp key
> ----------------------------------------------------------------------
> --
> To unsubscribe or change options: http://lists.samba.org/mailman/listinfo/rsync
> Before posting, read: http://www.tuxedo.org/~esr/faqs/smart-questions.html
>