How To Load An Activity In A fragment?

What are good interarrival and phase times for load testing?

• I'm using Tsung on an Amazon EC2 instance that is load testing our platform API. I'm using this currently:    <load>       <arrivalphase phase="1" duration="30" unit="second">         <users interarrival="0.125" unit="second"></users>       </arrivalphase>       <arrivalphase phase="2" duration="30" unit="minute">         <users interarrival="1" unit="second"></users>       </arrivalphase>     </load> I know every situation is different, but how about for an API that's being accessed by an iPhone/Android/iPad/Webapp that has about 4 million users, with activity around pro basketball games?

• Answer:

  All depends on the load you are expecting to see in practice. The Fukushima nuclear power plant designers weren't stupid. They had been told to design their plant to withstand waves up to 5.7 meters, and so they did, and as far as anyone could tell, the plant was able to withstand such waves just fine. The 2011 tsunami produced a 13 meter wave and the rest is history. That does not mean the plant wasn't designed to spec, but rather that the spec was overly optimistic. So too with software. A load test gives you an assurance that if your real load goes up to the level of the test load, your system will at least not fall over in a trivial way. (There could always be some way in which the real load is different from the test load even though it has about the same volume, and it is always possible that the system is somehow sensitive to that difference. But at least you know that if it falls over under a load equivalent to what you successfully tested, it's for some nontrivial reason.) Now what load you want to test for depends on what load you expect in practice. It's hard and possibly misleading to think of a number that gives you the absolute maximum load you'll see in real life. Load on your system is a random variable with a very smeared out probability distribution (most likely), and trying to determine its "maximum" involves making an arbitrary decision about where you cut the distribution off. Or rather, the decision is not entirely arbitrary: it should depend on just how bad the disaster would be if the system did fall over at some point. Don't take "never ever" for a criterion: that's pointy-haired-boss-speak. In the real world, there are tradeoffs. If you're building a bikeshed, you might not care if it floods once a year; bikes are made of steel, after all. If you're building a nuclear plant, you'd be well advised to make sure it's not going to melt down on average more than once every, oh, say, 1000 years. A highway bridge should probably be somewhere in between. (In decision theory jargon, your decision should depend both on the distribution of your random variables and on your cost function.) What kind of application are you testing? A bikeshed, or a nuclear plant?