Keep traffic distribution throughout ramp phase

[Alex Rousskov]

On 08/16/2013 09:34 AM, unjc email wrote:

> I started with populus_factor first; I found the webpolygraph client
> process did not start the robots with respect to the address
> distribution specified.  For example, I found no GET requests in the
> first 5 to 10 minutes of the test, it seems to run the POST robots first.  


When populus factor is used, the next robot to start is selected at
random. The only way I can explain the above anomaly is if your total
number of robots is so small that 60%*16% = 10% of them is still likely
to be zero.

  60% - proportion of GET ribits
  16% - 10 minutes out of 60 minute ramp phase

How many robots do you have total?


> Next I tried load_factor and found GET requests dominate the whole
> channel in the first 10 minutes while there is no QUERY or POST requests.  

With load factor, Polygraph starts all robots at the beginning of the
test. If you see no QUERY requests during the first 10 minutes, then
perhaps your peak load is so small that 30%*16% = 5% of that peak
results in very infrequent requests?

What is your peak request rate?


Perhaps you can post the entire workload?


Thank you,

Alex.


> [ GET.addresses: 60%, QUERY.addresses: 30%, POST.addresses: 10%] =
> robotAddrs(authAddrScheme, theBench);
> 
> Phase RampPhase = {
>     name = "Ramping";
>     goal.duration = 60min;
>     populus_factor_beg = smallFactor;
>     //populus_factor_beg = 0;
>     //populus_factor_end = 1.0;
>     load_factor_beg = 0;
>     load_factor_end = 1.0;
> };