This December Kernl launched it’s new WordPress Load Testing service. As part of the bug shakedown we decided to load test as many managed WordPress providers as we could. In this test, we turn our sights to WordPress.com.
How do we judge the platform?
For this series of blog posts we judge the platform via 3 different load tests.
- Baseline – This test is for 200 concurrent users, for 10 minutes, with a 2 user / second ramp up. We use this test to double-check our configuration before throwing heavier load at the provider.
- Sustained Traffic – 2000 concurrent users, for 2 hours, ramping up at 2 users / second. This test represents what a high traffic WordPress site might see on a day to day basis.
- Traffic Spike – The traffic spike test simulates what might happen if a Twitter or Instagram celebrity mentioned your site. 20,000 concurrent users, for 1 hour, ramping up at 10 users/s.
For all 3 tests traffic is generated out of Digital Ocean‘s San Francisco 2 (SFO2) data center.
What WordPress.com plan was used?
For this test we used the Free WordPress.com plan. We didn’t need any of the bells and whistles, plus (as you’ll see later) the performance didn’t suffer at all. If performance had been impacted we would have increased our plan to somewhere around the $10/month mark. As with all of our load tests we don’t do any configuration. We simply import the content of http://www.re-cycledair.com and then start testing.
The Baseline Test
200 concurrent users, 2 users / s ramp up, 10 minutes, SFO
As expected WordPress.com did very well during the baseline test. The test settled in at 26 req/s.
There also weren’t any failures through the duration of the test.
The response time distribution was also excellent with 99% of requests returning in under 200ms.
The Sustained Traffic Test
2000 concurrent users, 2 users / s ramp up, 2 hours, SFO
The sustained traffic test is where WordPress.com’s hosting really started to shine. As of now it is hands down the best host that we’ve tested. The throughput settled in at around 258 requests / second. There also weren’t any surprises on our way up to that number of requests.
The most impressive part about this entire load test was the failure rate. Over a 2 hour test, under heavy load, for more than 1.4 million requests, not a single request failed. Thats some serious stability.
While not as impressive as the 0% failure rate, the response time distribution was still pretty amazing. 99% of all requests finished in well under 100ms. There was an outlier in the ~1500ms range, but that isn’t uncommon for load tests.
The Traffic Spike Test
20000 concurrent users, 10 users / s ramp up, 1 hour, SFO
WordPress.com is blazing fast. It didn’t event flinch with 20000 concurrent users. The request rate settled in at 1717 requests / second (!). On the way up to that request rate there were no surprises or stutter steps.
The failure rate was exceptional as well. For an hour long test, with sustained heavy load, and a total of 4.3 million requests, there were 0 errors.
Finally, the most impressive graph in this entire test! For the traffic spike test, WordPress.com’s distribution chart is nothing short of fantastic. 99% of traffic had response times below 50ms, and even the 100% outlier was still only 1 second. Great work WordPress.com team!
Conclusions
If you need extremely robust performance and are OK with the restrictions of WordPress.com they seem like a great choice.
Want to be part of the Kernl WordPress Load Testing Beta? Sign up and then send an email to jack@kernl.us