Just wondering if anyone else has been experience flakey results with the Dulles site? I’ve been testing a bunch over the last several days and getting poor results compared to using other test sites.
I’m kind of a newbie, so not sure about all this, but mainly have been using Dulles until this week.
Here’s my situation. I run a bunch of sites for local businesses, mostly static Wordpress sites, pretty low volume. I use a VPS on Wired Tree, which until now has been quite good. I usually just optimize once and then test periodically to make sure my server is running fine. Most of my sites are pretty fast.
So last week or so I noticed some slow performance for the first time. Checked with my host and they said things looked good. I usually do the full 9 runs. This time, I started seeing various slowdowns, at least 3-4 run with TTFB of more than 1 sec (usually 200-300ms), or really slow response times for a couple file, but not always the same ones. Also, the first run usually seemed ok, I think mostly the consequent runs were slow. I saw this on several different sites of mine.
It was so unusual I decided to try other test locations, which I’ve never done. Turns out Lincoln, San Jose, Denver, all seem to give my typically “fast” results, so it seems like only Dulles is showing the slowdown. Any help would be appreciated. Here is a sample of some of the results:
One key difference is the are multiple agents in Dulles, so the tests may run in parallel whereas most of the other locations have one maybe two agents.
Here’s two tests I did first one where all the tests ran on the same agent (so ran in series), second where it used multiple agents so they ran in parallel
Although there’s some longer times in the first test, there’s no where near as many in the second test and from this I’d lean towards your server setup being the issues i.e. it slows down under multiple simultaneous requests.
You can check the server response times by adding the response time field (%T) to your Apache logs (http://httpd.apache.org/docs/current/mod/mod_log_config.html)
Thanks for that info, very helpful and makes a lot of sense. Just curious, how did you run the test using the same agent, I didn’t see any setting for that. Also, is the info available for the different locations as to how many agents they have?
I don’t know anything about Apache logs but I’ll certainly dig into that so I can check server response times. I want to have hard evidence to show my hosting company, otherwise they keep citing gtmetrix and the single-run webpagetest to claim my server is fine.
Finally, I did notice a “gap” in the waterfall view of some test runs and don’t know what that means; is it just waiting for server response? Sometimes it’s a small gap, sometimes as much as 2 secs, at different points, but usually in the bottom half of the waterfall, I’ve attached an image of one example.
To make sure all the tests ran on the same agent I ‘hacked’ the affinity field (https://sites.google.com/a/webpagetest.org/docs/advanced-features/webpagetest-restful-apis) into the submission form using DevTools
If you need to re-run the test you can just resubmit my test above.
If you want to verify this in something else then you could use the free version of Load Impact - https://loadimpact.com/
I’ll have a look at the gaps in the waterfalls a bit later.
Thanks again, Andy. So testing more of my sites gives similar results, clear slowdowns when testing from Dulles/multiple agents.
Any thoughts on what kind of settings on the server should be tweaked? I don’t really know anything about servers, but my host support techs keep insisting there’s no issues with the server. If I could point them in right direction it might help.
The techs keep giving me pointers on how to optimize my sites; I can certainly improve the sites, but they already have a fair bit of optimization and run pretty fast on single testing, so that should indicate it’s a server problem, no?
Or maybe I just need to find a new host, though I’d much rather not do that. Wired Tree has been very good up till now, this is my first real problem.