Guus Bosman

software engineering director


You are here

guusbosmannl

Performance tests Guus Bosman.nl

Yesterday at the end of the afternoon Mark and I did an experiment with my website.

We were looking at a program that can be used to measure the performance of a website: how many visitors per minute can it reasonably handle? We need such tools for our work, and we did a quick experiment with a new program that Mark found.

I don't know exactly how many visitors my website usually has. I should really find that out once! A very rough estimate (for last Tuesday) indicates that every two minutes a page of our combined websites gets visited.

Background
The website you are reading is running on a new PC, and that computer doesn't do a lot more than handling the website. That means that in theory it could deal with quite a lot of visitors at the same time.

However, the connection of my computer to the internet is a major bottleneck. It's an ADSL connection with a maximum upstream of about 380 Kbit/s. If there are a few people on my website this is definitely fast enough. More than, say, ten visitors per minute would be a problem: the website would become very slow.

The experiment
The program we used yesterday is called JMeter. It behaves exactly as a user visiting a website: the website doesn't see the difference between a real visitor and JMeter. Because it's automatic, you can configure it to behave as a group of visitors too; which is of course the powerful aspect of such programs. We configured the tool such that 40 people would visit my site every minute. Half of them would see the whole website (including pictures); the other half without pictures. We then run the test for 15 minutes.

The result
The main result: when 40 people per minute are visiting the site, it will take around half a minute for a new visitor before he sees the frontpage of the site, and another minute before all the images and pictures will have been downloaded. That's pretty good! Of course for a professional website that would be quite slow, but for my self-made and self-managed hosting solution it's quite reasonable.

Visitor Average time
With images 93 s
Without images 27 s
Total 60 s

We also re-run the tests with an even bigger amount of visitors. If 400 visitors per minute visit the site, many of them will be blocked: the computer simply refuses to serve them. This type of test is called stress tests: seeing if the server would crash under heavy pressure. With 100 visitors per minute the site gets very slow of course: an average of more than one minute per page without images, and much longer if images are included. But the important result is that the computer didn't crash, and kept functioning.

Line chart with the result of the performance test.

Comments

xs4all. will upgrade the speed of the adsl this month.

I know -- my downstream connection speed has been doubled this week!

I was very happy when they announced that ;)

Some remarks are to make:
- upstream speed is even "worse", and maximum 320 Kbit/s
- This was a stress test, in real life people read the text when they request a page. Normal behavior is a reading time of 20 sec. Two things can be done: another, more realistic test with waiting time or somne math... (I prefer the first one :-)

Yeah, a more realistic test could be interesting. Also with logged in users, and with users reading article-pages (not only the frontpage).

I liked the tool a lot, especially because it's so easy to set up. No scripting!

Hey Guus,

Interessante gegevens, ik host mn server gewoon hier onder de bank (samen met de ups) in de woonkamer; we hebben een roadrunner abonnement in chapel hill die het (geloof expliciet) verbied, maar ik heb nog nooit iets van ze gehoord.
Ik genereer te weining verkeer gok ik (ook nooit naar gekeken).
Anyways, ik zal die jmeter eens proberen, zien wat hier de bottleneck is.

mn web-server is 2.13Ghz, 512MB. hij runt een vm met daarin de e-mailserver, en is ook muziek-speler en smb file-server. Ik heb er een wiki en mijn en Quinn's wordpress blogs op.
Ben wel nieuwsgierig naar je ervaringen met drupal, en eventuele andere cms'en die je hebt gepreobeert.

Het enige echte nadeel van het thuis runnen is dat mn dynamic ip door veel mailervers hier is geblacklist, dus moet ik een externe smtp server gebruiken voor uitgaande mail.

maarten

ps als je het leuk vind moeten we eens afspreken, we wonen (in amerikaanse begrippen) practisch naast elkaar :)

Recent comments

Recently read

Books I've recently read: