Life in the (not-so) Fast Lane - Part II

Life in the (not-so) Fast Lane - Part II

Following on from my last post -- Life in the (not-so) Fast Lane -- I spent some time thinking a little more about the Internet response times we see here in Australia and decided to get some real stats to see how things really stack up out in the wild.

Thanks to the fine folk at WebPageTest.org, and armed with a list of the top 100 online retail websites from Alexa.com, I ran some web response time tests across them, hosted out of Sydney.

I tested the load times for the top 100 online retail sites (as of July 2015) on the webpagetest Sydney host, simulating a desktop computer using a cabled connection (5Mbps down, 1Mbps up, 28ms Round Trip Time with 0% packet loss) using the IE 10 browser.

In my hypothetical example previously, I used Australia's published average broadband connection speed of 0.94Mbp. I theorised that Aussie Joe Average should see a web response time of approximately 1.5s from the top 100 global websites (with an average 1.4Mb page size).

Using IE 10 on a 5Mbps download pipe sees the top 100 (retail) sites average 1.59s TTFB (time to first byte response) which is pretty close to what I thought. From there though, things start to go downhill, as the new figures I've tested also include client-side processing:

On average, 3.96s after sending the page load request, IE 10 starts to render the page. Page load completion times average 9.98s (this is the browser on-load window event), and the full page load time (including concurrent additional resource requests) averages 12.18s after the initial request. This is way, way worse than I originally thought things would be.

Ok, so this is the average across the top 100 retail sites. And averages usually fall foul of any poor performers, with stats being negatively skewed. So let's take another look at the numbers across the ten best-performing sites (returning the lowest TTFB response).

Time to first byte for these averaged 0.65s, with the browser render start time averaging at 2.48s. The on-load event averaged 6.36s with the full page load at 6.60s.

So, whichever way we slice these figures, -- using the best performers on a much fatter pipe than Aussie Joe Average has on his home broadband -- response times are around the 2.5s mark when rendering starts. And then it's going to be another couple of seconds after that before the pages render enough to do anything useful with them.

Admittedly, the results I'm posting here are nowhere near definitive due to (at least) the small sample size and short duration. But if nothing else, they've confirmed in my mind we've still got a very long way to go if we're ever going to break the 100ms speed record for instantaneous response here in Australia.

要查看或添加评论,请登录

Chris Jones的更多文章

  • Performance is not horsepower alone

    Performance is not horsepower alone

    I’m always banging on about performance being fit-for-purpose rather than just outright speed. At AccessHQ, we’re…

    3 条评论
  • Why do we never learn in IT?

    Why do we never learn in IT?

    I honestly can’t think of any other major industry which consistently over-spends, under-delivers and repeats the same…

  • Think you'd make a good Performance Tester?

    Think you'd make a good Performance Tester?

    “Some men aren't looking for anything logical, like money. They can't be bought, bullied, reasoned, or negotiated with.

  • Performance Is Not Just Speed

    Performance Is Not Just Speed

    A lot of people think performance in IT systems is all about speed. Questions like; “How fast does it go?” and “What’s…

    2 条评论
  • Performance Testing - are you peering into darkness?

    Performance Testing - are you peering into darkness?

    Within big IT projects, often performance tests run with few infrastructure monitors in place, if at all. It's not…

    2 条评论
  • Performance Testing Averages, 90th percentiles or Avg-90%?

    Performance Testing Averages, 90th percentiles or Avg-90%?

    As a performance tester, my role is to ensure my clients clearly understand how their systems perform under load. To…

  • Performance is not just reliability and availability

    Performance is not just reliability and availability

    There's no truer maxims in the world of complex IT systems than Murphy's Law, and its corollary Finagle's Law. These…

  • Why Testing is like Book Publishing

    Why Testing is like Book Publishing

    Testing software is a complex and difficult thing. There are so many opportunities for issues to arise - from major…

  • Performance Testing Cheatsheet - Diagnosing Server Congestion

    Performance Testing Cheatsheet - Diagnosing Server Congestion

    When faced with diagnosing performance issues with windows-based servers (especially when perfmon stats are easily…

    1 条评论
  • Open All Hours. 23? x 7; 357 Days a Year

    Open All Hours. 23? x 7; 357 Days a Year

    In today's online market, to coin a phrase, time is money. So, availability and performance are king, right? Well…

社区洞察

其他会员也浏览了