On Sat, Apr 5, 2008 at 7:16 AM, <[EMAIL PROTECTED]> wrote: > Thank Jae, Sebb, David for your response. > Jae/David, I am using only single JMeter instance (no master/slave > style) to run this 10 user test. > Sebb, I agree, I have one text response assertions, which I don't think > should affect transaction response-time. It should calculate the > response-time from the time it has got the response from server and > should not include response assertion duration in the transaction time. > And, delay in transaction submission is expected in this case (causing > tps to bring down). > Sebb, can you please explain how the connection handling is different in > AB & Jmeter? > Few more details: > No synchronizing of users in JMeter for concurrency (it is simultaneous > user-load scenario). > Data sent over are of same size in both; in fact it is xml payload for > web-service which is common. > > Warm regards, > Kaushal Kishore >
JMeter has a good HTTP stack, whereas AB does not. If you need to learn a real web server processing delay (subtract from it your network delay, which is normally constant, unless the network is congested), you need to test it using a client with a HTTP stack. Ether find a way to fine tune JMeter for that this purpose, or you may try curl-loader, which several people are claiming was good for that. Be cautions and measure (e.g. by ping), that you have constant RTT, and no network congestion issues at your load scenario. -- Sincerely, Robert Iakobashvili