[httperf] Can httperf generate session whose length isexponentiallydistributed?

魏丫丫 wyy01@mails.tsinghua.edu.cn
Sun, 9 Mar 2003 16:22:28 +0800


Dear Martin Arlitt,
   Thank you for your reply. My cgi-script includes a circulation. such as:
                
         for ($count=1; $count <= 35; $count++) {
      							$i++;
         						for ($j=1; $j<=20; $j++) {}                
        						 }

   I do so because I want to generate a cpu-intensive workload. I know SSL workload is cpu-intensive.
I don't know this method can replace SSL workload generation.
   The other question is  how to generate SSL workload? I can't find any man page about httperf supporting SSL.
If you are convenient, thank you for telling me the uri for the newest man page of httperf. I think the server
need apache+openssl only and the client's request url is https://...  As if httperf supports SSL, then in such 
enviorment, the workload is SSL workload and cpu-intensive. Am I right?
   Thank you for your reply. I am extremely appreciated if you can help me.



Sincerely,

Weiyaya























	

======= 2003-03-04 07:31:00 您在来信中写道:=======

>Weiyaya
>
>in my opinion the output from the two tests is quite similar, considering
>that your Web server is extremely overloaded. for example, if we breakdown
>the 200 sessions from each test:
>
>Test 1
>------
>  successful sessions: 11
>server failure status: 83
>       client timeout: 106
>
>Test 2
>------
>  successful sessions: 23
>server failure status: 65
>       client timeout: 112
>
>obviously there is some variability, but I think that is to be expected
>given that your server is overloaded.
>
>if you look at the average response times, you will see that they are also
>quite similar:
>
>Test 1: 2254 ms
>Test 2: 2269 ms
>
>regarding your second question, I do not think that SOMAXCONN is a reason
>of overload.  I would investigate why Apache is taking so long to respond
>to each request (i.e., what is the cgi-script doing?).  The server
>machine itself does not seem to be overloaded, as the time to establish
>the TCP connections is quite low:
>
>Test 1: Connection time [ms]: connect 0.5
>Test 2: Connection time [ms]: connect 0.3
>
>Martin
>
>
>