When Google goes down – time we all did more web performance testing

Date: 16th September 2010
Author: Deri Jones

So even Google has outages.

Yesterday (September 15th) afternoon (UK time) a bunch of Google stuff disappeared – Search, YouTube etc.  Not so bad that everyone seemed to have noticed – but a few folks twittered.  (Update: 12 noon 16th September: No big bloggers have picked it up, so maybe it was more limited in scope).  But Google has crashed before.

Apparently not only did this stop people using Google’s own services (Google Docs etc).

It even caused other web sites to drop off the Internet too.

Likely because of Google Analytics or DoubleClick code in the page that stalled when it couldn’t reach Google.

It looks like a lot of normal websites which have Google hosted code embedded in their pages for ad-serving, tracking and other services are at risk of being down just because Google is.

Whilst some savvy users may have realised that you could get around this issue by turning off JavaScript in your browser – hey, the average web user won’t know that.

And anyway, lots of sites don’t work anymore without javascript – there’s simply too much dependency on AJAX.

In our web site testing – we’re seeing more and more clients depending totally on AJAX; and have come to us for help monitoring multi-page user  journeys because their old monitoring approach of hitting a few URLs one after the just don’t work – a Journey needs to follow the dynamic AJAX links to do what users really do.

Not sure what the root cause will turn out to be – to have pulled down several parts of Google it’ll have to be a central infrastructure kind of thing like a network routing goof – hard to imagine Google depend on just 1 network pipe!

So I guess this adds one more thing to be added to the performance testing list for any etailer or site that uses Google analytics or other services:  Run a test after disabling all URLs apart from your own in the test virtual browser: and see if the pages are slower / die more frequently than when testing the same User Journeys normally.

My colleague Gomez commented with a sigh

Web browser based testing just got a little more interesting

If you try it on your site, confidence from your powers that be that you’ve covered all the sensible bases will rise  – so it’s maybe a good idea to add it to your regular web load testing plan examples and standard comparisons on new code release checks – that is if you’ve a test framework set up for web browser user journey testing.

Yesterday (September 15th) afternoon (UK time) a bunch of Google stuff disappeared – Search, YouTube etc.  Not so bad that everyone seemed to have noticed – but a few folks twittered.  (Update: 12 noon 16th September: No big bloggers have picked it up, so maybe it was more limited in scope)

Apparently not only did this stop people using Google’s own services (Google Docs etc).

It even caused other web sites to drop off the Internet too.

Likely because of Google Analytics or DoubleClick code in the page that stalled when it couldn’t reach Google.

It looks like a lot of normal websites which have Google hosted code embedded in their pages for ad-serving, tracking and other services are at risk of being down just because Google is.

Whilst some savvy users may have realised that you could get around this issue by turning off JavaScript in your browser – hey, the average web user won’t know that.

And anyway, lots of sites don’t work anymore without javascript – there’s simply too much dependency on AJAX.

In our web site testing – we’re seeing more and more clients depending totally on AJAX; and have come to us for help monitoring multi-page user  journeys because their old monitoring approach of hitting a few URLs one after the just don’t work – a Journey needs to follow the dynamic AJAX links to do what users really do.

Not sure what the root cause will turn out to be – to have pulled down several parts of Google it’ll have to be a central infrastructure kind of thing like a network routing goof – hard to imagine Google depend on just 1 network pipe!

So I guess this adds one more thing to be added to the performance testing list for any etailer or site that uses Google analytics or other services:  Run a test after disabling all URLs apart from your own in the test virtual browser: and see if the pages are slower / die more frequently than when testing the same User Journeys normally.

My colleague Gomez commented with a sigh

Web browser based testing just got a little more interesting

If you try it on your site, confidence from your powers that be that you’ve covered all the sensible bases will rise  – so it’s maybe a good idea to add it to your regular web load testing plan examples and standard comparisons on new code release checks – that is if you’ve a test framework set up for web browser user journey testing.

Top