A Step by Step Guide to setting up an AutoScaling Private WebPageTest instance

If you have any interest in website performance optimisation, then you have undoubtebly heard of WebPageTest. Being able to test your websites from all over the world, on every major browser, on different operating systems, and even on physical mobile devices, is the greatest ever addition to a web performance engineer’s toolbox.

One small shelf of Pat Meenan's epic WebPageTest device lab

The sheer scale of WebPageTest, with test agents literally global (even in China!), of course means that queues for the popular locations can get quite long – not great when you’re in the middle of a performance debug session and need answers FAST!

Also since these test agents query your website from the public internet they won’t be able to hit internal systems – for example pre-production or QA, or even just a corporate intranet that isn’t accessible outside of a certain network.

In this article I’ll show you how to set up your very own private instance of WebPageTest in Amazon AWS with autoscaling test agents to keep costs down

Once you have this you can optionally extend the setup, for example:

  • Always-on test agent(s) – get your test result faster
  • Your own test agent within the China firewall – get realistic test results from the perspective of a user behind the Great Firewall
  • Automation and scripting – schedule regular tests
  • Reporting and visualisation – graph your tests to find trends

We’ll get on to those topics later, but first up let’s focus on building a solid WebPageTest foundation!

Getting familiar with AWS

You don’t need to be an AWS guru to get this all running. AWS can be confusing, however this setup only requires a few clicks throughout the AWS console. Make sure you have your own AWS account ready to go.

There are 2 or 3 main areas we will need to use in AWS for setting up our private WebPageTest instance:

  1. A user with permissions to create new EC2 instances which the WebPageTest server uses to spin up test agents, and optionally S3 permissions to archive the tests.
  2. A place to archive test results, freeing up space on the WebPageTest server itself (optional, but highly recommended – and cheap).
  3. A VM instance on which to host your WebPageTest server which orchestrates the whole process.

Let’s start off with –

1. Create a WebPageTest User

IAM is the AWS Identity and Access Management (aka IAM – "I am"…) area; we need to create a new programmatic user for the WebPageTest server to run under.

Log in to your AWS console and head over to IAM by finding it in the Services menu:


From there, click Add user, enter a name, and select Programmatic Access:

AWS IAM - creating wpt user

Now we need to Set permissions. Select Attach Existing Policies and search for and select "AmazonEC2FullAccess" (so it can start test agents)…

AWS IAM - setting permissions for EC2

… and search for and select "AmazonS3FullAccess" (to archive tests into S3).

AWS IAM - setting permissions for S3

The tests can easily fill up even a large EC2 volume quickly, so archiving to S3 is strongly recommended – S3 is super cheap, EBS volumes are not.

Archiving tests to S3 won’t change how WebPageTest looks and feels; it will fetch the zipped test from S3 on demand incredibly quickly.

Copy the AWS Access Key ID and Secret Access Key for this new user somewhere as we’ll need them in a moment.

AWS IAM - user created

2. Configure Test Storage in S3

EBS volumes (the hard drive on a EC2 instance) will fill up quickly and, although expanding them is possible, it isn’t easy and it can’t be reversed. S3 is the super cheap, almost limitless, alternative.

Head over to the S3 area of your AWS console – again, you can find this under the Services menu:


Give your bucket a unique name – your bucket name cannot already exist in the same region, even in another user’s AWS account – and set the region where you’ll be creating your WebPageTest server:

AWS S3 - create a WPT bucket

With that done, now you’re ready to create the server!

3. Setting up the WebPageTest server on EC2

Now let’s create the actual WebPageTest server itself; this is quite a long set of steps so get that coffee ready!

Head over to the EC2 area within the AWS console and select Launch Instance:


Search for "webpagetest" and select Community AMIs:

AWS EC2 - choose a WPT AMI

Select the top webpagetest-server result.

Choose instance size: t2.micro is ok to start with as you can always scale up if necessary, and t2.micro is currently free:

AWS EC2 - set AMI size

This is an important step: instead of launching the server and logging in to edit settings, we can actually define the settings in "user data" which is passed in to the instance at start-up.

If you prefer, you can connect to your WebPageTest instance after launching and configure these settings directly in the file /var /www /webpagetest /www /settings/ settings.ini; you can copy the sample and edit it.

Tap on Configure instance details, then Advanced, and paste in user data similar to this, filling in the blanks:

# User to start/stop agents and save tests to s3
ec2_key=<the IAM user access key ID>
ec2_secret=<the IAM user secret access key>

# Set the API key to use when programmatically enqueuing tests
# Can't think of one? check out http://new-guid.com/
api_key=<choose a super secret api key>

# show user timing marks in waterfalls - very handy

# better images for screenshots - you're using S3, right?
# so you have the storage!

# archiving to s3
archive_s3_key=<the IAM user key>
archive_s3_secret=<the IAM user secret>
archive_s3_bucket=<the WPT archive bucket>

# number of days to keep tests locally before archiving

# run archive script hourly automatically as agents poll for work

The full list of options that can be set are over on the github repo for WebPageTest, as the settings.ini.sample file; as mentioned earlier, you could skip the user data step and set these options in a ini file to be created in /var /www /webpagetest /www /settings /settings.ini

AWS EC2 - set user data

Now we need to make sure we can access the server over port 80; for this you need to select Configure security group, then add rule and choose HTTP:

AWS EC2 - set security groups

Alright! Ready to rock! Let’s hit Review and Launch and after you’ve selected or created a keypair (used to log in to the instance), and after it’s finished initialising we’ll be given a URL, where it says Public DNS:

AWS EC2 - starting up

Head over to that URL and you should see the familiar WebPageTest homepage:

AWS EC2 - WPT server running, submitting a test

Try it out – pop in a URL, select a location, and submit the test. If you then check the EC2 area of your AWS account, after a moment you’ll notice a new instance starting up called "WebPagetest Agent":

AWS EC2 - WPT test agent being created

A new test agent can take a few minutes to actually connect and start testing – sometimes up to 10 minutes – but once connected it’ll automatically pick up the enqueued test and run it:

AWS EC2 - submitted test being picked up

That was easy, right? The server will update itself from the WebPageTest github repo regularly, as will the test agents. Your tests will be automatically archived to, and retrieved from, S3.

AWS EC2 - test running

Your private WebPageTest foundation has been laid! We will build on this over a few more articles.

BONUS) Scaling Up and Staying There

By default, only one test agent per 100 tests will be created per region. One of the reasons I wanted a private WebPageTest instance was my impatience at queueing up to find out how slow my sites are. If you can throw money at the problem, then change this in your user data:

# This will create one new test agent for every 5 tests queued
# *up to the location max*:

# The default max per location is 1, so you need to override it per
# location to enable scale out:
# etc

If there are no more tests queued which that agent can pick up for an hour (or however long you want; an hour is the default) then it will be terminated. This means that you’ll have to wait for the agent to be recreated next time, but it also means you’re not paying for an EC2 instance that you’re not using; these agents should be c4.large, so they’re not free tier eligible.

To avoid this you can alter the user data (you’ll need to stop the WebPageTest server instance before you can edit it) to add in a line for each location you want to keep an agent always available, e.g.:


It will now scale down to a minimum of one test agent for each of the specified regions.


Wondering why your test isn’t starting? Getting impatient? Check /install to make sure you have green everywhere, except at the bottom where no test agents will exist (since they’re spun up on demand):

AWS debugging - install check

You can also check /getTesters.php?f=html to see what is connected:

AWS debugging - no test agents connected

The test agents are only created on demand, so let’s ensure there’s a test actually registered at /getLocations.php?f=html:

AWS debugging - confirming a test has been submitted for a location

You can also log in and check what’s happening on the server:

ssh -i "<your keypair>.pem" [email protected]<the url of your instance>


ssh -i "webpagetest.pem" [email protected]

Now let’s check nginx for requests to getwork.php – this is the URL that test agents poll to pick up the next test in their queue:

tail -f /var/log/nginx/access.log | grep getwork


[email protected]:~$ tail -f /var/log/nginx/access.log | grep getwork - - [25/Apr/2019:20:42:01 +0000] "GET /work/getwork.php HTTP/1.1" 200 5 "-" "Wget/1.15 (linux-gnu)" - - [25/Apr/2019:20:43:01 +0000] "GET /work/getwork.php HTTP/1.1" 200 5 "-" "Wget/1.15 (linux-gnu)" - - [25/Apr/2019:20:44:01 +0000] "GET /work/getwork.php HTTP/1.1" 200 5 "-" "Wget/1.15 (linux-gnu)" - - [25/Apr/2019:20:45:02 +0000] "GET /work/getwork.php HTTP/1.1" 200 5 "-" "Wget/1.15 (linux-gnu)"

hmm… not a lot happening there. That’s the server pinging itself for some reason; you can tell this since the requests are from "". Usually after a few more minutes you’ll see: - - [25/Apr/2019:20:45:37 +0000] "GET /work/getwork.php?f=json&shards=1&reboot=1&location=eu-west-1&pc=EC2AMAZ-CO5OM1I&key=4d446cb0d60d76dced79ffa39cf3c1e953db594b&ec2=i-0323a543b58e9ba00&ec2zone=eu-west-1a&version=190221.200223&screenwidth=1920&screenheight=1200&freedisk=7.464&upminutes=9 HTTP/1.1" 200 282 "-" "python-requests/2.21.0" - - [25/Apr/2019:20:46:01 +0000] "GET /work/getwork.php HTTP/1.1" 200 5 "-" "Wget/1.15 (linux-gnu)" - - [25/Apr/2019:20:46:39 +0000] "GET /work/getwork.php?f=json&shards=1&reboot=1&location=eu-west-1_IE11&pc=EC2AMAZ-CO5OM1I&key=4d446cb0d60d76dced79ffa39cf3c1e953db594b&ec2=i-0323a543b58e9ba00&ec2zone=eu-west-1a&version=190221.200223&screenwidth=1920&screenheight=1200&freedisk=7.404&upminutes=10 HTTP/1.1" 200 31 "-" "python-requests/2.21.0" - - [25/Apr/2019:20:46:39 +0000] "GET /work/getwork.php?f=json&shards=1&reboot=1&location=eu-west-1&pc=EC2AMAZ-CO5OM1I&key=4d446cb0d60d76dced79ffa39cf3c1e953db594b&ec2=i-0323a543b58e9ba00&ec2zone=eu-west-1a&version=190221.200223&screenwidth=1920&screenheight=1200&freedisk=7.404&upminutes=10 HTTP/1.1" 200 284 "-" "python-requests/2.21.0"

Notice the timestamps show there was nothing going on for over 3 minutes after I had started to get impatient; AWS EC2 test agents can take a while to wake up and connect, so bear this in mind.

AWS debugging - test agent connected

Interesting point: notice the querystring parameters in the request; the test agent informs the server a lot about itself, even including available disc space.

Things to check:

  • User Data – did you set the correct IAM details for EC2? If not, then the WebPageTest server will not be able to spin up the agents
  • IAM – did you give EC2 permissions to the IAM user you created? If not, same issue as above.
  • WPT logs – you can check the logs for issues in the WebPageTest logs, which can be found in /var/www/webpagetest/www/log/ (not /logs/, as this is where the submitted tests are logged, not errors).
  • nginx logs – you can check if the agents are able to connect to your the WebPageTest server at all.


Hopefully you followed along and successfully set up your own private WebPageTest instance, and can now queue up all the tests you like!

If you have issues, then hit me up on twitter or head over to the WebPageTest forums – seriously, given how many times the same question must be asked in there, the gang are exceptionally patient and helpful.

Good luck!

AI Awesomeness Part Deux! Microsoft Cognitive Services Speaker Identification

The Speaker Recognition API: AI Awesomeness

In a recent article I introduced Microsoft Cognitive Services’ Speaker Verification service, using a recording of a person repeating one of a set of key phrases to verify that user by their voiceprint.

The second main feature of the Speaker Recognition API is Speaker Identification, which can compare a piece of audio to a selection of voiceprints and tell you who was talking! For example, both Barclays and HSBC banks have investigated using passive speaker identification during customer support calls to give an added layer of user identification while you’re chatting to customer support. Or you could prime your profiles against all the speakers in a conference, and have their name automatically appear on screen when they’re talking in a panel discussion.

In this article I’m going to introduce you to the Speaker Identification API from the Cognitive Services and go through an example of using it for fun and profit! Though mainly fun.

Continue reading

AI Awesomeness! Microsoft Cognitive Services Speech Verification

AI Awesomeness: The Speaker Recognition API

Microsoft have been consistently ramping up their AI offerings over the past couple of years under the grouping of “Cognitive Services”. These include some incredible offerings as services for things that would have required a degree in Maths and a deep understanding of Python and R to achieve, such as image recognition, video analysis, speech synthesis, intent analysis, sentiment analysis and so much more.

I think it’s quite incredible to have the capability to ping an endpoint with an image and very quickly get a response containing a text description of the image contents. Surely we live in the future!

In this article I’m going to introduce you to the Cognitive Services, focus on the Speech Recognition ones, and implement a working example for Speaker Verification.

Continue reading

London Bot Framework Meetup Numero Four

On the 16th January I had the pleasure of hosting another London BotFramework meetup at the newly constructed event space in the Just Eat offices.

London BotFramework Meetup #4

They’ve joined three floors with a staircase, so attendees can have beers and pizza upstairs while the presenters sweat with the AV equipment downstairs!

There was a great turnout for this one, including the usual gang and a few new faces too.

Before I get started, in case you haven’t already seen it, you should totally subscribe to the weekly Artificially Intelligent newsletter that has the latest news in AI, Chatbots, and Speech and Image Recognition!
Go sign up for Artificially Intelligent!


Just want to get stuck in? Here’s the video; first half is Jimmy, second half is Jessica.


For this meetup we were fortunate enough to have the Engström MVP power team, Jessica and Jimmy, who were in town for NDC London and graced us with their presence.

1) Developing Cross Platform Bots: Jimmy Engström

Jimmy Engstrom - Cross Platform Bots
The first session included several fantastic live demos where Jimmy creates a simple chat bot and, with minimal development effort, gets it working on Alexa, Cortana, and Google Home!

(Rendering my own ingenious Alexa BotFramework hack from last year quite useless!)

During the day Jimmy Engström is a .NET developer and he does all the fun stuff during his spare time. He and his wife run a code intensive user group (Coding After Work) that focuses on helping participants with code and design problems, and a podcast with the same name. Jimmy can be found tweeting as @apeoholic

2) Conversational UX: Jessica Engström

Jessica Engstrom - Conversational UX
In the second half Jessica gave a great overview of creating a framework to ensure your bot – speech or text based – seems less, well, robotic!

Some great takeaways from this which can easily be applied to your next project.

Being a geek shows in all parts of Jessica Engström’s life, whether it be organizing hackathons, running a user group and a podcast with her husband, game nights (retro or VR/MR) with friends, just catching the latest superhero movie or speaking internationally at conferences.

Her favorite topics is UX/UI and Mixed reality and other futuristic tech. She’s a Windows Development MVP. Together with her husband she runs a company called “AZM dev” which is focused on HoloLens and Windows development.

Follow her exploits over on twitter as @grytlappen


The updated event space at Just Eat is great and gives better visibility of the sessions thanks to stadium seating at the back.

The sessions were insightful and overall I think this went well.

Here’s to the next one and don’t forget to join up (and actually attend when you RSVP… ahem…)

Image Placeholders: Do it right or don’t do it at all. Please.

Hello. I’m a grumpy old web dev. I’m still wasting valuable memory on things like the deprecated img element’s lowsrc attribute (bring it back!), the hacks needed to get a website looking acceptable in both Firefox 2.5 and IE5.5 and IE on Mac, and what “cards” and “decks” meant in WAP terminology.

Having this – possibly pointless – information to hand means I am constantly getting frustrated at supposed “breakthrough” approaches to web development and optimisation which seem to be adding complexity for the sake of it, sometimes apparently ignoring existing tech.

What’s more annoying is when a good approach to something is implemented so badly that it reflects poorly on the original concept. I’ve previously written about how abusing something clever like React results in an awful user experience.

Don’t get me wrong, I absolutely love new tech, new approaches, new thinking, new opinions. I’m just sometimes grumpy about it because these new things don’t suit my personal preferences. Hence this article! Wahey!

Continue reading

London Bot Framework Meetup the Third

Welcome to the Third London BotFramework Meetup! Here's the line up

On Wednesday 22nd November 2017 I had the pleasure of running the third London Bot Framework meetup at the lovely Just Eat office in central London. The offices have been recently upgraded and the new meetup space has a huge 9 screen display a multiple mic speaker system, including a fantastic CatchBox throwable mic for ensuring everyone hears the audience questions

It has been a year since the previous one (whoops) but it was great to see some familiar faces return in the attendees. I had forgotten how much fun it is to emcee an event like this! Maybe next time I’ll be sure to just emcee and not also commit presenting a session too.

Continue reading

The Tesco Mobile Website and The Importance of Device Testing

A constant passion of mine is efficiency: not being wasteful, repeating something until the process has been refined to the most effective, efficient, economical, form of the activity that is realistically achievable.

I’m not saying I always get it right, just that it’s frustrating when I see this not being done. Especially so when the opposite seems to be true, as if people are actively trying to make things as bad as possible.

Which brings me on the the current Tesco mobile website, the subject of this article, and of my dislike of the misuse of a particular form of web technology: client side rendering.

What follows is a mixture of web perf analysis and my own opinions and preferences. And you know what they say about opinions…

Client Side Rendering; What is it good for?

client side rendering frameworks

No, it’s not “absolutely nothing”! Angular, React, Vue; they all have their uses. They do a job, and in the most part they do it well.

The problem comes when developers treat every problem like something that can be solved with client side rendering.

Continue reading

Building your first Botframework based Cortana Skill

Hi. I'm Cortana.

At //BUILD 2017 Microsoft announced support for Cortana Skills and connecting a Cortana Skill into a Bot Framework chatbot; given the number of chatbots out there using Microsoft Bot Framework, this is an extremely exciting move.

In this article I’ll show you how to create your first Cortana Skill from a Bot Framework chatbot and make it talk!


If you’re not already familiar with Cortana, this is Microsoft’s “personal assistant” and is available on Windows 10 (version 1607 and above) and a couple of Windows phones (Lumia 950/950 XL), a standalone speaker – like an Amazon Echo – and a plethora of devices that can run the Cortana app, including iOS and Android and plenty of laptops.

Cortana all the things, Derrick.

You’re going to be seeing a lot more of this little box of tricks (“Bot” of tricks? Box of bots?.. hmm…), so you might as well get in on the act right now!

Continue reading

Involved in a startup? Read this!

Having been the VP of Engineering at a startup, I understand a lot of the challenges. The technical ones relating to the solution you think you need to build, more technical ones relating to the solutions the investors want you to build, the development process to best fit a rapidly changing product, team, requirements, and priorities, as well as managing the team through uncertain terrain.

They’re the fun ones. The easy ones! Especially given how talented my dev team was.

The founder had the difficult challenges; define a product that could be a success, iterate that idea based on extensive user testing, and most importantly, ensure there was funding.

Luckily, our founder was as talented at soliciting funds as we were at building epic tech!

If you are involved in a startup, perhaps Just Eat’s Accelerator programme can help with both types of challenge!

Continue reading

Receiving Images to Your Skype Botframework Bot (v2!)

If you’re getting a “403” HTTP error when attempting to receive an image sent to your Skype bot, and the previous use of message.ServiceUrl to create a ConnectorClient didn’t work, try this more verbose version which explicitly sets the authorization header:

byte[] data;

if (image.ContentUrl != null)
    using (var connectorClient 
        = new ConnectorClient(new Uri(message.ServiceUrl)))
        var token = 
            await (connectorClient.Credentials as MicrosoftAppCredentials)

        var uri = new Uri(image.ContentUrl);

        using (var httpClient = new HttpClient())
            if (uri.Host.EndsWith("skype.com") 
                && uri.Scheme == Uri.UriSchemeHttps)
                    .Authorization = 
                        new AuthenticationHeaderValue("Bearer", token);

                    .Add(new MediaTypeWithQualityHeaderValue("application/octet-stream"));

            // Get the image in a byte[] variable
            data = await httpClient.GetByteArrayAsync(uri);