Just for fun to see if its possible and how much I can squeeze out of it. Somebody else created a similar test (with a little less test logic than mine) using wrk, https://github.com/wg/wrk. I wanted to compare the throughput that can be generated with jmeter, over that generated with more lightweight tools.
The discussion is a little more complex than this. Its the result of experimenting over the last two years with setups that give the best results in the case of tests that need to generate huge amounts of dynamic test data. Also, I've avoided the distributed mode setup because the master is a single point of failure (i'm not 100% sure), while having independent jmeter instances is more flexible (if you prefer using the shell). But considering the improvements made in 2.9 (or was it 2.8?) to it, I will have to reconsider that in the future. Adrian S On Mon, Feb 25, 2013 at 3:16 PM, Oliver Lloyd <[email protected]>wrote: > Hi Adrian, out of interest, what made you opt for vertical scaling using > one machine? Did you try scaling horizontally using distributed mode (or > even just running isolated instances)? > > > On 25 Feb 2013, at 12:57, Adrian Speteanu <[email protected]> wrote: > > > Sorry about that, pressed send by mistake, I updated bellow the rest of > the > > message. > > > > On Mon, Feb 25, 2013 at 2:52 PM, Adrian Speteanu <[email protected] > >wrote: > > > >> Hi. > >> > >> The two questions are incompatible. And also number 2. creates problems > >> with number 1. Please read on the subject, it was already answered in > the > >> mailing list before. > >> > >> As for #1, let me give an example of a test I'm running at the moment > (but > >> its not standard and it depends on the setup and this particular test): > >> * test system: > >> - machine used to generate traffic: dual processor (6 physical > cores > >> each, 24 threads total) with 64Gb RAM, 15k RPM disks, running CentOS > linux > >> - jmeter 2.9 (also used 2.8 with comparable results) > >> * test requirements: > >> - generate 30 thousands requests per second > >> > > In order to generate this much traffic with one single machine, I have > > tested multiple setups and configs and eventually used 10-12 instances of > > JMeter, each spawning 3500 threads. The throughput is limited per > instance > > using the Constant throughput controller, so it reaches the desired > level. > > This setup is required in order to generate a stable and predictable > > throughput from JMeter. > > > > Obviously, monitoring client side (JMeter) is inexistent (apart from > > summary in console), unless I suspect fowl play. Monitoring is done at > > application level. JMeter just gives me the load the app under test > > requires. In order to add monitoring from JMeter, I have to write a jtl, > > which significantly reduces throughput that I can generate, so when I do > > use it, only one out of 10-12 instances writes to disk. > > > > Cheers, > > Adrian > > > >> > >> > >> On Sat, Feb 23, 2013 at 7:52 AM, Amit Kumar <[email protected]> > wrote: > >> > >>> Dear All: > >>> > >>> I have couple of simple queries, please confirm: > >>> > >>> 1. How much threads are supported by a single machine at a time (I > heard > >>> 300, right?)? > >>> 2. In Listeners(Table/Tree): Some requests appear with warning(red > >>> symbol), what does it mean? Does it mean that those requests are > failed? > >>> [Please refer screen shot]. > >>> > >>> -- > >>> Thanks and Regards, > >>> Amit > >>> > >>> > >>> --------------------------------------------------------------------- > >>> To unsubscribe, e-mail: [email protected] > >>> For additional commands, e-mail: [email protected] > >>> > >> > >> > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
