Thanks for all of the info. And yes, the goal with the test client config was to try to basically guarantee a max out of the bandwith from any one client, then run multiple clients, all in an effort to swamp the whole connection at the server, and see what happens to the server -- hence the relatively unrealisticallly high load levels. The idea is to see how the server performs at its maxed-out bandwidth. The only issue we're having is that we can't run it long enough to really get a good steady state going given the mem issue.
As for that, hooking the two streams together should still be ok, as long as they get cleaned up at some point afterward... Not sure how we'd be getting an out of mem on this over time unless something is leaking somewhere. I will try and debug some more, but curious if this was relatively well-known. As far as the more general info, sounds like there are some interesting articles, must have missed part of the site, will check them out -- thanks. Will On 11/28/05, Peter Lin <[EMAIL PROTECTED]> wrote: > > other people have also tried large files in the past and jmeter definitely > has a limit. the current implementation simply passes the inputstream from > the file to the output stream to http connection. > > beyond that, unless a server is hosted at a tier 1 or tier 2 ISP, chances > are the connection won't be able to handle more than a dozen concurrent > uploads. the best option is to use multiple clients to send large > uploads. > though keep in mind a 100mbit ethernet can only handle 9-10Mb/second. > > 10mbit of bandwidth can only handle 1Mbyte/second. Many 2nd and 3rd tier > ISP > only give account 10mbit of bandwidth. Getting a full 100mbit bandwidth > generally only happens at tier 1 ISP like Level3, MCI, Global Crossing, > etc. > hope that helps. I cover this stuff in my articles on jmeter's articles > page. > > peter lin > > > On 11/28/05, Will Meyer <[EMAIL PROTECTED]> wrote: > > > > Hi. I have a test that, among many other things, uploads a huge file > via > > a > > POST. I have all of the listeners disabled, and am running the test in > > command-line mode. > > > > Eventually, with high enough numbers of loops and threads, I run out of > > memory. I have set the max heap quite high. Does anyone know what the > > memory implications in JMeter are of having a big (1M) file upload in > the > > post? Does it keep the data in memory for some period of time? > > > > Related, is it correct to assume that, with no listeners, the memory > > consumption of the process will not grow beyond that required (all else > > being equal) to hold tests for all the threads at one time, and that > there > > is no reason it should grow beyond that point? In other words, anything > > that would cause memory to grow over time that I could take a look > at? I > > would have thought that once I got to a steady-state, memory would spie > up > > and down as the tests are run and the GC runs, but the overall trend > > should > > be flat. > > > > Thanks, > > > > Will > > > > > >
