I had an interesting project – to say the least. All my attempts of trying to explain it to anyone have not gone too well for me. Therefore, I will just write a short summary of the essence of the project here. The following description comes from my Project Report that I had to submit at the end of my contract period (just because I am too lazy to write the same thing again in different words 😛 ).
The experiments at CERN generate mountainous amounts of data. Up to 40 Terabits per second amount of data more is expected to be produced during the future run of LHC in 2020. To filter the data coming from the LHCb detector, LHCb team is working on a benchmark application called DAQPIPE which emulates the data coming from the detector in a local area network. One of the major challenges faced by DAQPIPE is to prevent any type of data loss by implementing high-speed networks which can bear the load of the data coming in. The aim of the project is to implement a DPDK driver for DAQPIPE to create a high speed packet processing, transfer and communication mechanism for future LHCb upgrades.
DPDK is a new frame-work for very fast software defined networking. It allows multi-stage Ethernet networks for DAQ to be implemented with very cost-effective hardware by offloading all intelligence and most of the buffering into commodity servers. The LHCb data acquisition for Run3 will need a 40 Terabit/s network, where 100 Gigabit Ethernet is one of the interesting candidates. In this project we aim to port the existing DAQ exerciser software of LHCb, DAQPIPE, to DPDK and perform tests on state of the art network hardware.
LHCb Group Presentation: http://goo.gl/Xji9ix
Openlab Lightning Talk Presentation: http://goo.gl/UyDxDb