WebPagetest Forums

Hey guys,
as I'm writing on a new blog post, I've been doing some tests on some more or less know German shops from the different european test locations (except Dublin and Gloucester, because they don't support IE8 and I wanted consistent test conditions).
The result has been, well, unexpected. In general I'd assume that, the further the test locations get away from the shop's server the slower the time to first byte and load times will be due to latency. But this isn't always the case compared to the "home location" Frankfurt. I'll post the test results below including traceroute tests (wanted to be sure, they are really further away - at least by means of hops) and hope for your ideas. My only explanation so far would be very different internet connections to the test servers. The test conditions have been:

• IE8
• DSL speed
• 5 test runs
• first view only
  

Tests were run on http://www.strenesse.com/ which is hosted in Hannover, in the middle of Germany. But that's just an example.

Amsterdam, NL (11 hops):

• First Byte: 1.632s
• Start Render: 5.216s
• Load Time: 7.854s
  

Brussels, BE (14 hops):

• First Byte: 1.244s
• Start Render: 10.941s
• Load Time: 18.880s
  

Frankfurt, DE (8 hops):

• First Byte: 0.650s
• Start Render: 4.691s
• Load Time: 7.472s
  

Geneva, CH (11 hops):

• First Byte: 0.594s
• Start Render: 5.370s
• Load Time: 8.095s
  

London, UK (10 hops):

• First Byte: 0.561s
• Start Render: 4.028s
• Load Time: 6.434s
  

Moscow, RU (19 hops):

• First Byte: 0.676s
• Start Render: 5.610s
• Load Time: 8.765s
  

Paris, FR (14 hops):

• First Byte: 0.806s
• Start Render: 4.214s
• Load Time: 6.311s
  

Stockholm, SE (16 hops):

• First Byte: 0.584s
• Start Render: 4.051s
• Load Time: 6.370s
  

What surprises me most, are the results of London, Paris and Stockholm (and actually even Moscow for being that far away).


Thanks for any hints,
Martin

It looks like the first byte times for that site are largely a function of the server back-end performance which is quite variable. The socket connect times will give you the RTT and that is really the only part that varies by distance.

If you have a completely static html file served by a fast host (for consistent results) then you should see the kind of behavior you were expecting.

Another great case of not doing premature optimization - measure, see where the bottlenecks are and then address them. In this case the application back-end needs quite a bit of work.

Additionally you should have a look at the CPU Utilization. Especially Frankfurt node seems to be quite frequently short on that. Especially if you turn Video Capturing on. As soon as it hits 100% it starts to blur the picture.
Of course it tells you the real user-experience (of THAT node), but not the correct Network / Server Performance.
For this reason I moved as a default test node from Frankfurt to Paris. Even though Frankfurt is the closest, Paris performs much faster due to apparantly a higher CPU performance headroom.

Hope you'll point us to your blogpost here, as soon as it is published.

Kind regards,
Markus

Thanks for the hint with CPU usage - I'll use Paris for standard tests from now on. Seems like Brussels has the same Problem as Frankfurt... Too little CPU headroom.

You'll find the blog post at http://www.shopanalyst.de/2011/10/lange-...um-kunden/
It's in German, though the table comparing average Initial Connection times over some Shops hosted in Germany seen from the different European test locations should be readable for all.

Martin