On the second week of shopping, Performance is still quite good…
Stress Out Your Drupal Website with a World-class Specialist
The team over at Promet Source is sharing their expertise in a webinar showing you how to properly stress test your Drupal site. Andy Kucharski has asked me to join him in sharing best practices with anyone needing to understand the load testing process.
Register for the webinar
Andy is a frequent speaker at Drupal camps and conferences all over the country. He is recognized as one of the top minds in the world when it comes to architecting an effective Drupal implementation. Andy's reputation recently has skyrocketed as the foremost performance specialist on Drupal technology.
One of the coolest parts of the webinar will be Andy's data that compares the performance of Drupal hosted on Amazon versus Rackspace. His team has done some real load testing with Drupal identical applications installed on different environments. There are statistics for AWS EC2 Small and Medium instances, and there are stats for Rackspace Cloud 2G and 4G instances.
The metrics obtained by Andy and his team are quite interesting to performance geeks like us. Don't you want to know how much horsepower translates to what percentage of faster response times?
Andy shares their findings for number of peak users supported, as well as the experiment parameters, graphs, etc.
It should be a fun time. For us geeks obsessed about speed, scalability, and performance optimization. We expect plenty of Drupal experts to attend as they learn some extremely useful information regarding the best hosting options.
Register for the webinar
See you on Tuesday!
How to Manage User Experience for Silverlight and Flash Applications
Server metrics feature released
Load Impact is proud to present our latest feature - server metrics! With the server metrics feature you will be able to see performance metrics from your own servers along with the regular Load Impact metrics. These include, but are not limited to, CPU usage, memory usage, network activity and disk usage. This will allow you to correlate the different metrics and make it really easy to pinpoint the bottlenecks of your site.
To use the server metrics feature you will need to install a small program, what we call the server metrics agent, on the server you would like to see metrics from. Once installed, the agents will appear on your test configuration page in a new section. Choose the ones you wish to use in the test and run the test as you normally would. The metrics can then be plotted as graphs in the result page, just like any other result data.
The server metrics feature is in beta, but available for everyone, so there might be a few bumps that we need to flatten out. We will hopefully catch most of them over the next few days with the help of you all. Initially we have set a limit to 3 server metrics agents per test. This limit might be adjusted later. If you need more server metrics agents in the same test, feel free to contact us.
If your server is a Windows server, we provide an installer with server metrics agent and everything needed to run it. For our Ubuntu users, we provide a repository and a deb package. For everyone else, we provide a tar file. The source code for the server metrics agent will be put on GitHub for anyone to see and read. If you are hesitant to installing software on your server, then just go ahead and read the source code and you will be able to tell exactly what the agent does, and perhaps more important - does not, do.
So, what do you need to do to get started with server metrics testing? The first thing you should do is to generate a server metrics token. This token is used to identify your agents when they are talking to Load Impact. You generate the token on your profile page. The next step would be to download and install the server metrics agent for your server platform. Once the server metrics agent is configured you select them in your test configurations. Please refer to install instructions in the README file for details.
Load Testing to Prepare for 2 Million Pageviews
Yesterday I received an email from a client named Mike. He's a nurse. He is not a web developer, nor is he a software tester. He runs a very successful Christmas website TheSantaTracker.
Anyway, he wrote a few paragraphs on his perspective (inexperienced, non-technical) on load testing. I thought I would post it here to share with everyone - especially because he has nice things to say about us! ;-)
Some of this will sound like a commercial, and I apologize in advance for that. However, I hope those readers who are not regular coders or quality assurance folks will find these "rookie load tester" thoughts helpful. It goes to show that not everything has to be overly complicated.
Sometimes simple solutions are the best solutions.
Mike took a good bit of energy to write this, and I'm going to share it. Also, I want to support him in any way I can because he brings happiness to children that are in tough times. Have you ever had a kid in ICU? I have - and it is very hard to deal with. The nurses make all the difference in the world. They are heroes, and they make a difference every single day on the job.
I salute Mike Walczak for everything he does - both in the ICU and with his Christmas website that helps people.
"As a pediatric ICU nurse, expecting the unexpected is something that I experience every shift. In order to deal with the high level of uncertainty that occurs in the workplace, I find it especially important to be prepared. Preparation may sound simple, but it often involves taking many factors into consideration. As time consuming as preparation may be, it ultimately grants a person the peace of mind he or she needs to know that they are ready to handle anything.
"When I started my Christmas website over 4 years ago, I did it with the intention of bringing happiness to some of the kids in my family. However, since my initial launch, the website has skyrocketed beyond the s cope of a simple family tradition. Last year, the site served over 2 million pages on Christmas Eve alone. With the increasing traffic also comes the interesting predicament of finding the right type of Hivelocity Hosting server hardware to support the annually growing audience. This year, I stumbled upon the great services at LoadStorm.com and quickly realized that this may be the perfect solution to test my new server setup.
"Now let me first point out, I am certainly no webserver junkie nor do I have any experience in load testing webservers. Despite my lack of experience, I was able to quickly understand the necessary steps to setup a test load. I first took a look at how much the site has grown over the past few years in order to gauge an idea of how many users I should realistically test for. Also, recalling that preparation is key, I factored in reasonable “worst case scenario” type numbers. When I finished
running the numbers, I came up with the right number of concurrent users to test the server with. (By the way, I also found a 2010 LoadStorm blog post that helped break down how to come up with the right amount of concurrent users for your own websites.)
"For the second step, I looked into access logs which revealed some of the heaviest
used pages on the website. Given that users were using certain pages more than others, I wanted to ensure that my load test took these pages into consideration. LoadStorm is fantastic in that it actually allows you to setup a custom scenario where you can specify which pages are viewed and in what order (of course including a field wildcard steps along the way). The scenario process is extremely quick and it even shows you a preview of the content that the load servers will see during the test (it is helpful to know what exactly will be displayed).
"Finally, I setup a duration period for my test. I decided to run my test over an hour so that both my hosting provider and myself could monitor how the server setup was handling the load. Loadstorm even has real time performance metrics that help you to understand how the server is responding from a user perspective. This ultimately was a wonderful feature since we did in fact decide to tweak several settings early on that definitely had a great impact down later in the test. Not to mention, it was great to be able to find and tweak these settings now instead of when the website is actually receiving a lot of real traffic.
"At the end of the day, the entire process was painless and efficient. The server handled the excessive traffic without any major issues and I now have the piece of mind that the server is ready for the annual Christmas audience (plus extra). It is also worth mentioning that although I had thousands of concurrent users being sent to my server from only a handful of IPs, the system never caused me or my DDOS system any issues. I’d like to thank Scott and the entire LoadStorm team for creating such a fantastic product; it really is extremely easy to use. I would also like to thank the team at Hivelocity Hosting for providing me with such incredible hardware this year. Without a doubt, your products are world class!"
What's New | December 2012
Over the past several months, we have been gathering your feedback through survey results, reading your discussion posts on Community, and listening to our support teams. We take your comments to heart and do our best to help make your product experience a great one.
With our most recent product release, we covered a full range of UI/UX enhancements, new features, and bug fixes. See a full listing of these features below, and again, please complete our survey or chat with us on Community if you have any additional product enhancement ideas you’d like to share!
Overall User Experience Design Enhancements
Our new look offers a fresh and simplified perspective on how you view all of your accounts and services, including monitoring, load testing, and/or real user measurements activity.
Heading over to Scripting, you will notice overall User Interface enhancements with a new look to the Overview page and the addition of your list of Scripts. You can now easily search your scripts, upload or create a new script, and re-validate a script all from the same page.
Share your monitoring and real user measurement results - without the requirement for your recipients to log in!
To share graphs, first select the monitors you want to graph, then hit the Graph button. Once you are at this window, you can now share the graph with your stakeholders without having them log in to the account. Simply copy and paste the URL to share with them when the pop-up window opens.
Here's the view in Monitoring:
Here's the view in Real User Measurements:
More Features for Real User Measurements and Monitoring
Not only can you share the time series graph with your stakeholders, the time range has been enhanced to include 90 days and 1 year.
Additional data to help make quick decisions
Group by and filter JS errors: Use the new filtering functionality to quickly find out which are the most common JS errors on your website, and when each error occurs (either per URL - see screenshot, or per browser, country, etc.). Note that we only capture JS errors happening up to 1 second after the browser's onload event. Those errors are the most likely to affect your users' experience when the page is loaded.
Web Beacon: You may now select Maximum Load Time and Collection Sampling Preferences to improve your results
For additional information, check out the latest post from the lead developer of this product.
Invoice Billing versus Credit Card Purchases
For larger purchases, we have streamlined the process to allow you to request for invoice billing during checkout rather than paying by credit card.
Real User Measurements - End of November release
We are excited to announce the latest features of our Real User Measurements product.
UI - Time Series
You can now share the Time Series graph with anyone. Also, the time range has been enhanced to 90 days and 1 year.
UI - Analysis
You can now group by and filter JS errors. This allows you to quickly find out which are the most common JS errors on your website, and when each error occurs (either per URL - see screenshot, or per browser, country, etc.). Note that we only capture JS errors happening up to 1 second after the browser's onload event. Those errors are the most likely to affect your users' experience when the page is loaded.
UI - Web Beacon
Two preferences have been added:
- Define the maximum load time allowed in a measurement (default is 300 seconds). Depending on how long your page load time tends to be, this will help exclude outliers that might skew your average load time. But do not set it too low or you may exclude measurements that are worth keeping.
- Set the collection sampling rate (default is 100%). This is completely optional. At this point we recommend that you keep this setting at 100%.
RUM web beacons call our servers using the same protocol (http or https) as your web site's URL. It used to be https only.
Holiday 2012 – In the first week of the season, my site performance was…
Scheduled Maintenance for 12/01/12 - 10 PM PT - 3 AM PT
We will be performing maintenance to our Community site on Saturday, 12/1 between 10PM PT to Sunday, 12/2 3AM PT.
During this maintenance window, there should be no impact to you but if there is, please contact us at .
Cyber Monday Wrap-up: The Power of Positive Performance
about:performance is changing…
Midday Cyber Monday – The Bad Taste of Third Party Performance
Thanksgiving Weekend Analysis – Big Online Sales, but No Performance Surprises
Slides from Agile Testing Days Berlin
Black Friday 5:00 PM – Busy But Uneventful – A Day Without Performance Drama
Black Friday Noon Update – A Few Hiccups But No Major Outages
Black Friday Morning – All is Quiet…So Far (November 23 2012 – 09:00 EST)
How Will Retailers’ Sites Perform During “Make it or Break it” 2012 Holiday Season?
Load generation from Australia
A few days ago Amazon announced the availability of its new AWS data center in Sydney, Australia. We here at Load Impact are now happy to announce that we have implemented support for load tests from the Sydney data center.
This means that we are now able to generate simulated user traffic from 8 different locations around the globe:
- Palo Alto, California, USA
- Portland, Oregon, USA
- Ashburn, Virginia, USA
- Sao Paolo, Brazil
- Dublin, Ireland
- Singapore, Singapore
- Tokyo, Japan
- Sydney, Australia
You are of course also able to use multiple locations where your traffic is generated, in a single load test. This is something unique to Load Impact that other services don't offer.
Some other new things we have released recently include:
- Parallel deployment of test configurations for large tests - This will drastically reduce the time it takes to start larger load tests. Where previously we configured all involved load generator cloud instances in a load test sequentially, we now configure them in parallel.
- Graceful test shutdown - Previously, when a test was completed, the load generator instances would be killed abruptly. This meant that tests usually ended at the exact moment they were supposed to, but it also meant that some transactions that were "in transit" would never be recorded by Load Impact, even though they might appear in the logs on the web server. To make this less likely to happen, we have now started shutting down tests more gracefully than before, waiting a little bit for all simulated clients to be done with their current transactions. This means that while earlier, all clients in a test would just disappear when the test ended, you will now see a short ramp-down period where the number of clients ramps down quickly from whatever level you were at, to zero.
- Various bug fixes and improvements - We have implemented lots of smaller bug fixes and improvements, for example in script validation, script conversion, user session handling, etc.