[OpenSER-Users] UDP Buffer Size / OpenSER / Linux

[Michael Young]

I've been working with this also, and I don't understand the decision
process that OpenSER uses to select its UDP buffer size.

1. The system default is 128 Kb. When I start OpenSER without changing any
settings, it uses a 128 Kb buffer size.

2. I changed the OS buffer to 256 kb (sysctl -w net.core.rmem_max=262144).
OpenSER starts, and says:

INFO:core:probe_max_receive_buffer: using a UDP receive buffer of 255 kb

OK, so far so good.

3. So next I set the OS buffer to 1 Mb (sysctl -w
net.core.rmem_max=1048576). I did not set the MAX_RECV_BUFFER_SIZE or use
-b. Openser starts and says: 

INFO:core:probe_max_receive_buffer: using a UDP receive buffer of 512 kb

Hmmm, that's strange....

4. I start Openser with -b 1048576. Openser says:
INFO:core:probe_max_receive_buffer: using a UDP receive buffer of 2048 kb

2 Mb? The OS limit is 1 Mb...  That can't be right.

5. I've been fiddling with various numbers for -b, and if I use 1024,
Openser says: 
INFO:core:probe_max_receive_buffer: using a UDP receive buffer of 124 kb

6. If I use -b 1024000, Openser says:
INFO:core:probe_max_receive_buffer: using a UDP receive buffer of 2000 kb


So now I'm confused. I can't get Openser to give me a UDP buffer size of 1
Mb, it seems to settle on either 512 Kb or 2 Mb. Am I missing something?
This is with 1.3.2.

Michael Young


-----Original Message-----
From: users-bounces at lists.openser.org
[mailto:users-bounces at lists.openser.org] On Behalf Of Bogdan-Andrei Iancu
Sent: Wednesday, April 09, 2008 7:37 AM
To: Carsten Bock
Cc: users at lists.openser.org
Subject: Re: [OpenSER-Users] UDP Buffer Size / OpenSER / Linux

Hi Carsten,

MAX_RECV_BUFFER_SIZE (or -b cl param) is used only as upper limit when auto
discovery/setting the socket buffer. 


Regards,
Bogdan

Carsten Bock wrote:
> Hi Bogdan,
>
> i have another question regarding the UDP buffer size:
> If we set the UDP-Buffer of the system to, let's say 1 MByte: Do we also
> have to edit MAX_RECV_BUFFER_SIZE? What happens, if we don't? Looking at
> udp_server.c it seems like this value is ignored...
>
> Thank you,
> Carsten
>
>
> Am Dienstag, den 25.03.2008, 12:53 +0200 schrieb Bogdan-Andrei Iancu:
>   
>> Hi Klaus, Hi Carsten,
>>
>> Setting a large buffer for the UDP socket will not solve the problem of 
>> slow openser processing, but will solve the problem of temporary 
>> overloads due a short traffic peek or due short openser blocking in an 
>> synchronous op (like DNS or DB).  In such corner cases, the buffering in 
>> the kernel socket will avoid message dropping.
>>
>> Regards,
>> Bogdan
>>
>> Klaus Darilion wrote:
>>     
>>> Carsten Bock schrieb:
>>>   
>>>       
>>>> Hi,
>>>>
>>>> short question:
>>>> Can anybody give me a hint regarding the Linux UDP-Buffer size? Are
>>>> there any known good values for this or is the default value (128k) of
>>>> Linux sufficient? I've read, regarding other SIP-Systems (namely IBM
SIP
>>>> Servlets and Communigate pro), the recommendation to increase the
>>>> UDP-Buffer-size to 3 or even 4 MByte....
>>>> Are there any recommended / known best values for heavy load OpenSER
>>>> Installations?
>>>>     
>>>>         
>>> IMO these values are too high. If the server is for any reason to slow 
>>> to handle the load and the buffer fills up, then it might happen that 
>>> the transactions are already timeout in the UAC when the SIP proxy 
>>> starts processing the requests. If the proxy is overloaded IMO the 
>>> requests should be dropped instead of queuing them for too long time.
>>>
>>> regards
>>> klaus
>>>
>>> _______________________________________________
>>> Users mailing list
>>> Users at lists.openser.org
>>> http://lists.openser.org/cgi-bin/mailman/listinfo/users
>>>
>>>   
>>>       
>
>
>   


_______________________________________________
Users mailing list
Users at lists.openser.org
http://lists.openser.org/cgi-bin/mailman/listinfo/users

No virus found in this incoming message.
Checked by AVG. 
Version: 7.5.519 / Virus Database: 269.22.9/1365 - Release Date: 4/8/2008
7:30 AM
 

No virus found in this outgoing message.
Checked by AVG. 
Version: 7.5.524 / Virus Database: 269.23.21/1456 - Release Date: 5/20/2008
6:45 AM