[OpenSIPS-Devel] New OpenSIPS design and load balancing

[Aron Rosenberg]

We have been doing a lot of R+D work on making OpenSIPS work in a Load
Balancer environment and some of the issues that we are facing would be
interesting to design into the new core.

The usage of Shared Memory to cache data in various modules makes it
difficult to scale these modules to run on several physically separate
nodes. We have contemplated using DB only mode, but the performance of
this is subpar and rapidly becomes a DB scaling issue.

One of the items that might be interesting to think about would be a new
class of memory/object storage that I will call "Global Memory". Global
Memory could be a key/value storage system for data which modules need
to create/access/modify/update that is Dialog level rather than
transactional level. Global Memory could be backed by something like
memcached.

Most modules that build internal caching lists from shared memory could
be converted to this new system. Usrloc, registrar, presence.

Shared memory would then be used for single node only processing such as
transactional memory, or local copies of module data that does not
change. The load balancers that we are working with provide node level
SIP transaction persistence.

-Aron

Aron Rosenberg
SightSpeed Inc.

-----Original Message-----
From: devel-bounces at lists.opensips.org
[mailto:devel-bounces at lists.opensips.org] On Behalf Of Bogdan-Andrei
Iancu
Sent: Friday, November 07, 2008 4:32 AM
To: users at lists.opensips.org; devel at lists.opensips.org
Subject: [OpenSIPS-Devel] RFC: new opensips design

Hi all,

As previously stated, there is a common consent that the current 
design/architecture of OpenSIPS/OpenSER (inherited from SER) is no 
longer able to deliver and to meet the present requirements and demands 
for OpenSIPS future evolving. It is simply an inevitable dead-end that 
needs to be avoided.

Why? - I made a summary here - 
http://www.opensips.org/pmwiki.php?n=Development.NewDesign#toc1

I guess all of you are aware of the critical problems like blocking DB 
or RADIUS, blocking DNS, lack of scaling, too complicated scripting, 
message changes processing , etc.

To be able to deliver a solution that will be able to satisfy the the 
growing complexity and scale of the SIP world, a new radical design is 
needed. It will be process of intensive thinking and work, but the 
result will be a license for the future.

As a first step of this laborious process, collecting feedback from all 
of you, about the is missing, what are your wishes, about the 
known/unknown limitations, etc I just open an new web section where we 
can compile all this information.

It is not a request for solution, but a request for current problems! We

need to know all the problems that people facing in order to come up 
with a design that will solve them all.

Please visit:
    http://www.opensips.org/pmwiki.php?n=Development.NewDesign

and freely post there - any feedback is a valuable input and investment 
in this project.

Thanks you,
Bogdan


_______________________________________________
Devel mailing list
Devel at lists.opensips.org
http://lists.opensips.org/cgi-bin/mailman/listinfo/devel