Only one, similar to UDP.

Regards,

Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
http://www.opensips-solutions.com

On 16.09.2014 13:03, george wu wrote:
Does SCTP need only one socket for 100K end points?
Can it be secure?  If so, that's really great.

George Wu



At 2014-09-16 05:54:42, "Bogdan-Andrei Iancu" <[email protected]> wrote:

    SCTP ?

    Bogdan-Andrei Iancu
    OpenSIPS Founder and Developer
    http://www.opensips-solutions.com

    On 16.09.2014 12:41, george wu wrote:
    Bogdan-Andrei Iancu:

    One more question, I know there is tls for tcp, is there any
    similar for udp so that
    the udp package is secure? Thanks.

    George






    At 2014-09-16 03:34:27, "Bogdan-Andrei Iancu"
    <[email protected]> wrote:

        Hi George,

        There are advantages and disadvantages for each protocol you use:

        UDP is much lighter as protocol for the OS (100K UDP end
        points -> only one socket, 100K TCP end points -> 100K
        sockets) and also for the application (OpenSIPS) - managing
        TCP connection is more resource intensive rather than UDP sockets

        UDP also have some advantages as it a framed protocol - a SIP
        packages and delimited at protocol level (in a single
        datagram), while in TCP, where everything is streamed, the
        reading application cannot "see" at transport level where a
        SIP package end and a new one starts - it has to do SIP
        parsing just to figure out where it ends.

        Nevertheless, TCP has better NAT penetration (as it is
        connection oriented) - this is why all mobile devices do
        prefer TCP over UDP.
        Also UDP has problems when comes to size, as it is limited to
        65K (not to mention poor UDP fragmentation on several stacks)
        - again, TCP solves this problem as it is stream oriented.

        Best regards,

        Bogdan-Andrei Iancu
        OpenSIPS Founder and Developer
        http://www.opensips-solutions.com

        On 12.09.2014 17:13, george wu wrote:
        I am not sure if I should start another thread or not.
        I have similar question.
        The client is android linphone. it says it needs to send
        keep-alive message for firewall traversal.
        For tcp it only needs to send every 10 minutes while for udp
        it needs to send every 10 seconds.
        It is obviously tcp is much better for the clients.
        However from thread here, it is better to use udp.

        Now can anybody give me some clue how to do tradeoff between
        tcp and udp.
        Thanks.

        George





        At 2014-09-12 09:38:04, "Bogdan-Andrei Iancu"
        <[email protected]> wrote:

            Hi Jayesh,

            There are 2 aspects here :
                - first, configure proper limits when starting
            OpenSIPS - like max number of fds per process, etc.
                - secondly, when comes to opensips itself, you need
            to look into :
                    * enough memory (TCP uses a lot)
                    * set proper timeouts in TCP (connect, write,
            read timeouts) to avoid blocking
                    * properly handle the TCP lifetime to get to a
            compromise between the number of ongoing connections and
            seting/closing connections
                    * really good control over when OpenSIPS should
            open new TCP conns - you can do this from script,
            depending on the target (like never try to open conns
            towards end-user, let them connect to you).

            Regards,

            Bogdan-Andrei Iancu
            OpenSIPS Founder and Developer
            http://www.opensips-solutions.com

            On 10.09.2014 14:55, Jayesh Nambiar wrote:
            Hello,
            I am in process of designing opensips which can handle
            a million users, hypothetically 200,000 registrations
            and 500CPS capacity. I've been reading a lot and learnt
            that TCP design is blocking and not suitable for huge
            volumes.
            My requirement was to have TLS between endpoints and
            Opensips and hence riding over TCP is the only option.
            I needed some expert suggestions on what things should
            be taken care of when planning a large deployment over TCP.
            I have built similar stuff on UDP and I trust it
            heavily as it has performed perfectly as expected. But
            when it comes to TCP I'm a bit clueless.
            I've read about lot of global parameters that is now
            available for asynchrous tcp operations. Is it really
            helpful when planning for scalable environments??

            Thanks for any suggestions !!

            W/regards,
            Jayesh


            _______________________________________________
            Users mailing list
            [email protected]
            http://lists.opensips.org/cgi-bin/mailman/listinfo/users





        _______________________________________________
        Users mailing list
        [email protected]
        http://lists.opensips.org/cgi-bin/mailman/listinfo/users







_______________________________________________
Users mailing list
[email protected]
http://lists.opensips.org/cgi-bin/mailman/listinfo/users

Reply via email to