[tahoe-dev] atlasgrid performance testing results

[Brian Warner]

On 9/23/11 6:33 PM, Zooko O'Whielacronx wrote:
> Could you provide us with the New Visualizer display of a small-K
> immutable download and of a large-K immutable download so that we can
> compare the two?

Hm, that's an interesting question. I could provide a screen capture of
such a display, and if you were lucky, I might have zoomed it in to a
useful region of interest. But since it's really a zoomable/pannable
interactive display, the real way to hand you such a display would be to
give you the JSON dataset and the javascript program to render it.

Which makes me think of the "tahoe web-app" idea I've pondered in the
past, and whether it might be interesting to have content-renderers
built in to tahoe. I'll write it up in a different message, though.

> And could you likewise do runs of those two cases with the "--profile"
> option to "tahoe start"?

I can try.. it'll be interesting to see how that affects the timing.

> Or, even better, let's automate all of the above and have buildbot do
> it regularly from now on. :-)

Hm, interesting. For that performance report, I ran a random-download
tool continuously for about 24 hours per graph, collecting a few
thousand samples each. The visualizer display would only be run on a
single download, so I'm not sure it's really useful unless/until we see
some other performance problem and want to learn more about what's
happening. An automated --profile run.. now *that*'s intriguing.. sort
of like the automated code-coverage tests, something to tell us what
code is on the critical path at any given time. Hmmmm.

cheers,
 -Brian