RestcommONE jSS7 performance exceeds 24000 MAP messages per sec!

SS7 as protocol is the lowest layer in over all PSTN world and should be able to handle messages from millions of subscribers. RestcommONE jSS7 stands to the challenge and provided a very robust, scalable and open source SS7 stack. Recently SS7 team did lot of work on improving the performance. One of the bottleneck identified was transportation layer – SCTP. jSS7 SCTP layer was built using the JDK’s com.sun.nio.sctp.* package that utilizes native lksctp library of linux. jSS7 SCTP layer utilized NIO and single thread model for IO and another thread for delegating the data to higher layer. While this was able to take load to certain extent, however at heavy loads the performance degraded and hence TeleStax jSS7 team decided to move to Netty.

After using Netty to handle lower layers and improving on thread locks at higher layers – M3UA and TCAP, RestcommONE jSS7 is able to achieve 24000 map messages per second! For load testing, tool used was map load test built in-house. Tests were made on below server with both client and server running on same server

For Client and Server memory assigned to JVM was 2 GB each. Tests were done using JDK 1.7.0_75.

Below is the wireshark map statistics summary showing jSS7 stack achieving 24000 map messages per second!

The flow of test message was based on USSD messages exchanged between client and server. Below is the flow diagram.

The memory consumed was very stable at around 300MB. CPU usage arround 40%. Below is the snapshot from JConsole

All the projects are open source and available under AGPL. You can download and understand more about TeleStax’s SCTP layer from http://github.com/restcomm/sctp and jSS7 can be downloaded from http://github.com/restcomm/jss7

Feel free to do the load test locally in your environment and publish the results here or on forums.

If you like big challenges and are interested to help contributing to the effort, feel free to contact us to contribute the projects by signing our Contributor License Agreement at https://www.telestax.com/open-source/#Contribute !