Management, Optimization and Evolution of the LHCb Online Network

The LHCb experiment is one of the four large particle detectors operated at the Large Hadron Collider (LHC) at CERN. It is a forward single-arm spectrometer dedicated to test the Standard Model through precision measurements of Charge-Parity (CP) violation and rare decays in the b quark sector. The LHCb experiment will operate at a luminosity of 2 x 10$^{32}cm^{-2}s^{-1}$, the proton-proton bunch crossings rate will be approximately 10 MHz. To select the interesting events, a two-level trigger scheme is applied: the first level trigger (L0) and the high level trigger (HLT). The L0 trigger is implemented in custom hardware, while HLT is implemented in software running on the CPUs of the Event Filter Farm (EFF). The L0 trigger rate is limited to about 1 MHz, and the event size for each event is about 35 kByte. It is a big challenge to handle the resulting data rate (35GByte/s). The online system is a key part of the LHCb experiment, providing all the IT services. It consists of three major components: the Data Acquisition (DAQ) system, the Timing and Fast Control (TFC) system and the Experiment Control System (ECS). To provide the services, two large dedicated networks based on Gigabit Ethernets are deployed: one for DAQ and another one for ECS, which are collectively referred to as Online network. A large network needs detailed monitoring for its successful operation. Commercial network management systems are quite expensive an d difficult to integrate into the LHCb ECS. A custom network monitoring system has been implemented based on a Supervisory Control And Data Acquisition (SCADA) system called PVSS which is used by the LHCb ECS. It is a homogeneous part of the LHCb ECS. In this thesis, it is demonstrated how a large scale network can be monitored and managed using tools originally made for industrial supervisory control. The thesis is organized as follows: $\bf Chapter 1$ gives a brief introduction to LHC and the B physics on LHC, then describes all sub-detectors and the trigger and DAQ system of LHCb from design to performance. $\bf Chapter 2$ first introduces the LHCb Online system and the dataflow, then focuses on the Online network design and its optimization. $\bf Chapter 3$, the SCADA system PVSS is introduced briefly, then the architecture and implementation of the network monitoring system are described in detail, including the front-end processes, the data communication and the supervisory layer. $\bf Chapter 4$ first discusses the packet sampling theory and one of the packet sampling mechanisms: sFlow. It then demonstrates the applications of sFlow for the network trouble-shooting, the traffic monitoring and the anomaly detection. $\bf Chapter 5$, a possible upgrade of LHC and LHCb is introduced, the possible architecture of DAQ is discussed, and two candidate internetworking technologies (high speed Ethernet and InfiniBand) are compar ed in different aspects for DAQ. Three schemes based on 10 Gigabit Ethernet are presented and studied. $\bf Chapter 6$ is a general summary of the thesis.