Optimization of Data Transfer for 100 Gb/s Ethernet

In 2019 the LHCb experiment will go through an important upgrade, that will improve performance in many fields. One oh these fields is the DAQ system: it consists of a big flow of data that comes from the network and reaches a set of disks, where data are stored. This flow of data needs to be placed on the hardware in order to optimize the performance: high bandwidth and low CPU utilization are two key ingredients to achieve this goal. The first part of this project is about this: we realize a very basic test setup, consisting of two servers connected through a 100 Gbit/s Ethernet Interface and we optimize the data transfer in terms of bandwidth and CPU utilization.
In the second part a set of disks is connected to one of the two servers, in order to simulate a typical storage system. After optimizing the data transfer, a series of measurements of bandwidth as a function of the packet size is performed: the reason for this is to study the behaviour of the Data Direct Input/Output (DDIO) technology, that allows data packets of certain sizes to go directly from network to L3 cache, without touching the main memory; this should, in principle, enhance the performance in some cases.