Slashdot Mirror
Web Server Stress Testing : Tutorial /Review
darthcamaro writes "I found an interesting article on builder.com that suggests laying 'Siege' to your server to help you setup up a site to withstand /.
"One of the great fears for many Web developers, and even more so for Web server admins, is watching their sites brought down by an avalanche of traffic. The traffic may be from a DoS attack, or maybe the site just got slash-dotted. Bottom line: It just isn't available.""
Forward /. hordes! You have a mission now!
No, really, I don't think someone would try "frist pst0ing" this one...
No more trial and error for this guy. What I would do is purchase large amounts of traffic, tweak, then gradually work my way up to larger and larger numbers.
Not only does this save time, but hey, it saves money.
Just to clarify, I'm not encouraging anyone to dos the site or anything, just do what you normally would do without the parent post.
Would be rather comical for about 5 minutes if C|Net networks sites where brought down due to be /.'ed
GeekLeak.com - Silly name, serious geeks
$ siege -c25 -t1M www.mydomain.com
** Siege 2.59
** Preparing 25 concurrent users for battle.
The server is now under siege...
In other news, domain/hosting company mydomain.com was under a heavy DDoS attack, it's believed that the attacks were done by members of a geek news website called Slashdot.
The IT section color scheme sucks.
1. download ./configure
2. gunzip and untar
3.
4. make, make install
5. $ siege -c200 -t720M www.thescogroup.com
** Siege 2.59
** Preparing 200 concurrent users for battle.
The server is now under siege...
I looked through the article, it doesn't look like much more than a slightly sophisticated wget for loop :). Seriously though, this seems similar to a few other basic stress testers out there. For the projects I've worked on you need session management, interactive processes, ... basically hitting 5 urls isn't gonna stress test anything of value.
The Grinder on the other hand, allows for distributed workers, following the same or different 'scripts' all controlled from a single console. It provides you with a slew of configuration options and all sorts of data at your fingertips. The scripts are jython which is easy to learn and very flexible. If you want to stress test a complex app, especially something interactive, or requiring sessions, check out the grinder, it's a god send.
http://monkeyserver.com --- weeeeee
Another great tool for stress testing your site is Jakarta JMeter. Gives you a nice GUI for watching your response times plummet as your site is pummeled.
From the article:
Siege now supports a new proxy tool called Sproxy, which harvests URLs for testing. The premise behind Sproxy doesn't make much sense to me... Personally, I prefer Scout for getting my URLs, since it just goes through a site's links and adds them that way.
The advantage of using a browser to set up your test plan is that it better simulates real traffic patterns on your site. Microsoft's Application Test Center does this, and JMeter has a proxy server similar to Sproxy.
When you're trying to replicate problems with a live site, however, it would seem more appropriate to me if you could base your test on real traffic to the site. I wrote a load testing tool once that used web logs to simulate the actual traffic patterns, but it was incomplete, mostly because web logs don't record POST data. A good stress tool could come with an Apache/IIS/Tomcat plugin that recorded traffic for use in stress testing.
reech bee-yond ur clip-0n
Ok guys, time to confess. Who is sieging the RIAA website? It's been down for 5 days.
Why can't I load the article?
Yesterday was the time to do it right. Are we having a REVOLUTION yet?
stress testing based on concurrent users hitting script etc is fine,
but there are other things to make sure of, returnin information and the like
check out the perl module called webtest or something like that
back in the day we didnt have no old school
Any author who has any even remotely possibly interesting-to-geek material on their site who is not on a dedicated box with a full T3 who doesn't reject any request that is referer'ed by /. is behind the times anyway.
It is more productive to voice thoughtful opinions (reply) than to judge (moderate) others.
read the article. what is lacking from the article is consideration of business requirements and needs. Performance is always mediated by business requirements. Laying siege to a webserver tells you the breaking point, but more importantly than the breaking point is using stress testing as part of a deployment process and maintenance plan. Once you know the breaking, you still need to measure the optimal performance and have a plan for acquiring new hardware. Sudden spikes are unpredictable, but you can configure modern webservers to deny connections temporarily until the load returns to normal. People serious about maintaining 99.99% uptime run regular stress tests and have well document plans for acquiring new hardware.
One item this article didn't explicitly look at was the network saturation percentage.
Most servers can handle a greater load than the network traffic can handle. To demonstrate a proper test you would need to test outside of your routers and firewall. This means that the test machines should be located outside of your local area network while testing or at least a certain quantitative percentage for statistical purposes.
Odds are most people are going to work within the LAN and lay Siege to their machines but forget that there is an outside world.
I was expecting an article about stressing out a server by putting MS Exchange on it... But this is good too.
The best stress tester was a company called Envive, which was a distributed attack sort of focus, with server time and space all over the world. You write a script, and then can watch the attack from a web browser. Proof positive that Siege is more popular though - they went out of business.
/usr/bin/grep -i -E meaning life.txt
http://www.linux.ie/pipermail/ilug/2004-January/00 9863.html
Takes a bit to get into the discussion though
Relavent system has 2TB of data(6TB space), max recorded throughput is 550Mb/s, over 20000 concurrent http requests.
As a test consultant working in all areas of automated testing (as opposed to manual 'tick the box' testing) - I do most of my Load or Stress Testing using the industry standard tool LoadRunner. I've used all other load testing tools and this is by far the best (albeit pretty expensive) but for large scale commercial projects - nothing even comes close.
... but they couldn't sell it so they stopped development. MS Web Application Stress Tool "webtool" is worth a look. It free and does alot http://www.microsoft.com/technet/itsolutions/intra net/downloads/webstres.mspx