Benchmarking Ubuntu

[Randy Appleton]

On Tue, Sep 29, 2009 at 3:37 PM, Martin Olsson <martin at minimum.se> wrote:

> Randy Appleton wrote:
> > What tests would be most useful?  It has been suggested to me...
> >
> > - Time to boot
>
> I think this one is important, but it's also
> very carefully analyzed by many others already.
> You have seen this right?
> http://www.bootchart.org/images/bootchart.png
>

Yep.  And it's exactly the right tool.  I take suggestions for any more
tools too.


>
> > - Time to resume from sleep
>
> I think this one is the most interesting. In particular
> it seems that Apple has done a great job with this on
> their latest OS (in particular how fast they retrieve
> an DHCP ip address when resuming). Many Linux distros
> take a few seconds extra before you can actually use
> a web browser for example, after resume.
>
> > - Time to login
>
> Same as boot? However, time from login prompt to fully
> loaded desktop is more interesting. The latest Ubuntu
> version has something called sreadahead though which
> should probably not be counted towards the total time
> because the sreadahead process runs with very low CPU
> and IO priorities. But something like "from login prompt
> until all icons on the desktop is rendered" or "from login
> prompt until the Applications menu in GNOME is opened".
>

Yes, this is exactly what I mean.  Sorry I was unclear.




> It would also be interesting to see how these measurements
> different when certain input variables change. For example,
> when there is 1, 10, 100 or 1000 icons on the desktop,
> how does the corresponding time grow? linearly? quadratically?
> or even exponentially?
>

I don't know.  Does anyone really have 1000 icons on their desktop?  I could
do no icons and 20 icons.  That might be interesting.  Alternatively if
number of desktop icons has no real impact, then I report that.



>
> > - Time to find a file using  the default search tool
>
> Depends a lot on harddrive speed, number of and size of
> the documents, how they are organized into folders etc.
> I think it would be hard to get useful data out of this one.
>

It would also depend on which tool each distribution uses.  I *think* Sues
uses beagle and Ubuntu and Fedora uses trackerd.  If so, I would expect
performance differences.




> > - Time to load a document in OpenOffice
>
> This could turn out to be interesting if some input
> variable was modified during repeated benchmarks.
> For example, how does the load time grow when the
> number of pages the document has increases?
>
> > - Time to view a folder with 1,000 documents in it.
>
> This one is another good one. Here I also would like
> to know how the time grows for 1, 10, 100, 1000, 10000
> documents.
>

It occurs to me it would also matter if the folder has been previously
visited and had the icons made, or if it's a new folder (like a memorystick)
that needs to get it's icons made.


This could also be interesting, even though updating
> the distro is a much more rare task for the user (hopefully!)
>


Thanks for the suggestions.  I'd love to hear more suggestions from you (or
anyone else).

-Randy Appleton
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/ubuntu-devel-discuss/attachments/20090930/c9a4a5f4/attachment.html>