[AGENT++] Memory management of AGENT++

[wancheng82]

Apostolos,
Thank you.
You are right.
After removing the MAC, I find that arena is always equal to fordblks .

Robin



2010-07-28 



wancheng82 



发件人： Apostolos Manolitzas 
发送时间： 2010-07-27  17:05:44 
收件人： wancheng82 
抄送： agentpp 
主题： Re: [AGENT++] Memory management of AGENT++ 
 
On 07/27/2010 11:55 AM, wancheng82 wrote: 
I use AGENT++ in one open source project (www.pica8.org), and it is very convenient.
I test very large MAC address table in BRIDGE-MIB. 
At first it seems AGENT++ has memory leak int test 1.1) and 1.2), but when I do more test in 1.3) and 1.4), it suprises me.

Test 1)
1.1)Add 30000 MAC address to BRIDGE-MIB.
     The code :
    //MAC is in char buf[]
    if((row = dot1dTpFdbEntry::instance->find_index(buf)) == NULL)
       row = dot1dTpFdbEntry::instance->add_row(buf);
   The memory test:
    USER       PID  %CPU  %MEM    VSZ   RSS   TTY      STAT START   TIME COMMAND
    root     23482  11.2     2.2      72772 68976 pts/10   S+   16:43      0:02 ./agent   =====>about 72MB

1.2)Rrmove 30000 MAC address from BRIDGE-MIB
     The code:
     //MAC is in char buf[]
     if((row = dot1dTpFdbEntry::instance->find_index(buf)) != NULL)
        dot1dTpFdbEntry::instance->remove_row(buf);
    The memory test:
    USER       PID  %CPU  %MEM    VSZ   RSS   TTY      STAT START   TIME COMMAND
     root     23482  2.6        2.2  72772   68976 pts/10   S+    16:43    0:04 ./agent       ======>still about 72MB

1.3)Add another 30000 MAC address to BRIDGE-MIB(different from 1.1).
     The code :
    //MAC is in char buf[]
    if((row = dot1dTpFdbEntry::instance->find_index(buf)) == NULL)
       row = dot1dTpFdbEntry::instance->add_row(buf);
   The memory test:
    USER       PID  %CPU  %MEM    VSZ   RSS   TTY      STAT START   TIME COMMAND
    root       23482  1.6       2.2  72772 68984 pts/10       S+   16:43   0:06 ./agent   ======>72MB

1.4)Rrmove 30000 MAC address from BRIDGE-MIB
     The code:
     //MAC is in char buf[]
     if((row = dot1dTpFdbEntry::instance->find_index(buf)) != NULL)
        dot1dTpFdbEntry::instance->remove_row(buf);
    The memory test:
    USER       PID  %CPU  %MEM    VSZ   RSS   TTY      STAT START   TIME COMMAND
      root     23482  1.6        2.2  72772 68984 pts/10   S+      16:43   0:08 ./agent   ======>72MB

Question:
It seems that AGENT++ has its own memory management mechanism, is that right? 
How should I know them from the code?  I did not find them in code, because remove_row will delete the row. 

Thank you.
Robin
  

    Try to use struct mallinco [1] for actually metering the memory allocated and freed because Linux malloc implementation of stdlibc does some kind of caching in order to reuse the data. 
Especially check the fordblks part.

cheers,
-Apostolos

[1] http://www.gnu.org/s/libc/manual/html_node/Statistics-of-Malloc.html#Statistics-of-Malloc