← Back to team overview

launchpad-dev team mailing list archive

Re: Not a metric performance information

 

On 09-12-03 06:55 AM, Jonathan Lange wrote:
Hello all,

This isn't a metric as best as I can tell, but I thought you might be
interested.

I set up my personal server to fetch an arbitrary bug page from edge
Launchpad and time how long it took. I then graphed the data.

Points worth noting:
  * there's a lot of variation
  * it never goes below 2s

Attached is the picture and the script, in case you are interested.
I'm personally more interested in trends than the numbers themselves,
so I'm going to keep this running.


Interesting, thanks for the numbers.

I see a similar variance when loading a full page and all fifty or so of its resources: some inexplicably take longer than others to load, even thought they are similar in size. Reload the page, and the slow-to-load resources change. Reload again, and they change again.

I have a theory about the varying load times: there is at least one slow app server in the pool. If you load fifty resources from a pool of seven fast appservers and one slow one, then some of those resources may come from the slow server. Repeat the batch of fifty multiple times, and you start to see the random variance that I have.

It will be interesting to see the API numbers as well. That's important, as the API serves our AJAX page data.

It should be possible to put some hard server stats behind this theory, shouldn't it? Average response time per page served, grouped by page ID and app server, is one way to start.


Maris

Attachment: signature.asc
Description: OpenPGP digital signature


References