One reason I’m not blogging much is that I’m bleary-eyed from getting an all-new corporate Web site off the ground. Here’s the fun I’ve been having.
I spent yesterday beating on Web servers with load-testing software. I’ve got a couple of servers set up as candidates for our new corporate Web site.
Things were going great until I got to the second server. It produced tons of 404 errors. I found the problem and fixed it, but the errors never went away. A link checker confirmed everything was working, and I can navigate the site just fine in a browser.
After spending hours chasing the problem I decided the software must be pulling files from a cache. I turned off every cache I know about, but there could be a PHP cache out there I’m not aware of. I’m hoping the cache expired overnight. If not I’ll be chasing that down. So that’s what I’m doing this morning. In the afternoon I’m attending a training Webinar for the load-testing software.
RSS Posts Feed
I work on a software performance/stability team and have used all sorts of load testing software. If I was running my own shop, I’d ditch Mercury/HP load runner and stick with free stuff like JMeter.
There’s just not enough transparency from these big box vendors. There are some feature advantages, but few people actually use them all.
Good luck.
What all tools do you like?
This was the trial version of Neoload. Nice, but I couldn’t justify the expense of the unlocked version.
When I’m training people (usually developers) on this stuff, I generally recommend JMeter (http://jakarta.apache.org/jmeter/). It’s open source, and is pretty simple to get started with.
Driving load is not that complicated in itself. What is more important is to monitor the system under load during your test. Specifically things like: CPU, memory, and network I/O. That way, at a given concurrency and rate of request you can understand your system capacity. Unfortunately JMeter does not monitor your system – it simply applies load.
At what level of traffic will you need to upgrade your hardware and/or bandwidth?
If you’re spending money for a commercial tool, it should help you to answer that question.
For load testing, I wouldn’t worry a ton about a tool that can crawl all the links. What I would focus on are the heavy requests, such as searches, shopping cart operations, etc.
One thing you also want to keep in mind is being able to get a true outside perspective before launching an application to your user base.
Jmeter is great for inside the firewall testing initially, but you want to make sure that your application is also ready to go from the outside as well. There are a number of companies that offer this type of load testing.
In the past I have used Webmetrics
http://www.webmetrics.com/products/site_stress.html
and have been happy with their offering.
-Joel
Working at a company that offers load testing services, I’ll make the argument that the best use of your time is probably not in running load tests. Working with a managed load testing service not only helps you figure out the right way to test, and what to do with the data, but also allows you to spend your precious time on the things that set your business apart. Generally we find that our customers do a hybrid approach, where they run basic (small) tests themselves, and work with us to bring the big load (and expertice) at various milestones.
.-= Lenny Rachitsky´s last blog ..Press Release: GameStreamer Launches New Service with Neustar =-.