Hallo, A Correction in my mail >>This helped the packet loss to reduce by 70% but could not achieve 0% packet loss . This helped the packet loss to reduce by 10%. Before increasing buffers we have Faults rate of ~8 to 10% after increasing buffers we have fault rate of 0.9%. Thank you, Samba.
On Sat, Mar 19, 2011 at 12:13 PM, Sambasiva Rao MANCHILI < [email protected]> wrote: > Hallo Sip Experts / Transport Group Experts, > I am not a SIP expert of application layer. I am part of Test System > Development where we simulate the SIP traffic over different transport types > like UDP,TCP and SCTP. > When we are simulating just SIP signalling messages between our two test > systems at the rate of greater than 500 UDP packets per second, we > observed some of > Subscribers fail Signalling.After further investigations it is found that > UDP packets are lost consequently Signalling is failing for > some subscribers intermittently. > > To overcome this we then increased the udp.recvspace and udp.sendspace and > socketbuffer size in our Networking stack. > This helped the packet loss to reduce by 70% but could not achieve 0% > packet loss . > *Qeury 1*:- Is increasing buffer space is a solution or workaround for our > Network stack to overcome packet loss ? > > Default values of our Network stack were 40K for udp.recvspace and 9K for > udp.sendspace and socketbuffers are 256K. > We changed these values in our Network stack to 256K for udp.recvspace, > 256K for udp.sendspace and socketbuffers to 1024K. > This change has reduced packet loss but can could not make 0% packet loss. > > > Query2:- Can you please suggest us what should be the right values for such > buffer space if this is right approach ? > > We have yet to enable the voice transmission as a next immediate step, then > again SIP RTP UDP packets will further increase based on the codec chosen > and this will > cause even more packet loss. > > Query3:- What is the ultimate solution for eliminating packet loss both > on Signalling and on speech path ? > > Thank you in advance for your time. > Samba. >
_______________________________________________ Ietf mailing list [email protected] https://www.ietf.org/mailman/listinfo/ietf
