Difference in Autobench between num_conn and low_rate / high_rate?

Question

I'm running a load test using Autobench and httperf, and am having trouble understanding the example they give on their site. The example uses the flags

--low_rate 20 --high_rate 40 --rate_step 5 --num_call 10 --num_conn 5000

I get that this will start with 20 connections, and step up to 40 at a rate of 5 with 10 requests per connection. What I don't understand is what num_conn specifies. Are these like a "pool" of 5000 connections which the 20, 25, 30, etc. get pulled from? Or something else entirely?

Answer 1

Regardless of the rate, autobench will make num_conn connections per test. As rate increases, the duration per test decreases.

num_conn specifies the number of connections that will be made. (5000)

num_call specifies the number of calls (requests) per connection. (10)

rate specifies the number of connections per second. (20 initially)

So the total number of requests in a test is: num_conn * num_call (50000)

The duration of a test is: num_conn / rate (250 seconds initially)

The (attempted) requests per second in a test is: num_call * rate (200 initially)