[Kamailio-Devel] SF.net SVN: openser:[5764] trunk

[Ovidiu Sas]

On Thu, Mar 26, 2009 at 11:52 AM, Daniel-Constantin Mierla
<miconda at gmail.com> wrote:
>
>
> On 03/26/2009 05:46 PM, Ovidiu Sas wrote:
>>
>> On Thu, Mar 26, 2009 at 11:42 AM, Daniel-Constantin Mierla
>> <miconda at gmail.com> wrote:
>>
>>>
>>> On 03/26/2009 04:52 PM, Ovidiu Sas wrote:
>>>
>>>>
>>>> This new statistic is not a counter.  It is a dynamic statistic that
>>>> fluctuates in time and therefor it is not suitable to be exported like
>>>> a regular counter core statistic.  That's the reason for not adding it
>>>> to core_stats.[ch].
>>>>
>>>>
>>>>
>>>>>
>>>>> Just several statistics are exported by core (those prefixed with
>>>>> core::).
>>>>>
>>>>> You can move it better to statistics.{c,h} if these are new statistics
>>>>> exported by core, but I have not seen changes to core_stats.c -- where
>>>>> the
>>>>> table for core exported statistics is.
>>>>>
>>>>>
>>>>>
>>>>
>>>> <snip>
>>>>
>>>>
>>>>>
>>>>> Better let's find the proper place for this code and move it directly
>>>>> there,
>>>>> rather than simply revert. For example I would need access to cpuload
>>>>> details, code which is now in ratelimit module. Maybe we can combine
>>>>> the
>>>>> two
>>>>> in a single place accessible by the two modules.
>>>>>
>>>>>
>>>>
>>>> statistics.[ch] seems to be an ok place for now, but maybe it should
>>>> be better to create a new file for dynamic core statistics.
>>>>
>>>> The cpuload related code from the ratelimit has the same issue as the
>>>> code from snmpstats - it is not portable (it relays on linux specific
>>>> /proc directory).
>>>>
>>>>
>>>
>>> yes, I know, it is why I related both of them. IMO collecting these
>>> helper
>>> functions in new files for now (e.g., procinfo.[ch]) and move as lib in
>>> sr
>>> would be a solution. Then can be still exported as statistic by snmpstats
>>> and accessed as value by ratelimit.
>>>
>>> BTW, the /proc/net stats can deal only with tcp and upd? I haven't seen
>>> code
>>> about tls and sctp
>>>
>>
>> tls and sctp are still tcp connections, so tcp sockets are used.
>>
>
> right for tls, which comes over tcp, but sctp is a different transport
> layer, I don't think it maps statistics over tcp, but...

You are right about sctp.  Don't know how it is mapped in the /proc/net tree.

Regards,
Ovidiu Sas