Case Study: Performance Analysis of a Virtualized XRootD Frontend in Large-Scale WAN Transfers

TL;DR

This paper analyzes the performance of a virtualized XRootD storage frontend in large-scale WAN data transfers, demonstrating high throughput and system behavior under real production loads.

Contribution

It provides a detailed case study of a complex virtualized storage architecture's performance in high-throughput WAN transfers, including system configuration and observed results.

Findings

Achieved an aggregate throughput of 51.3 Gb/s under peak conditions

Observed data flow peaks of 41.5 Gb/s validated by external tools

Demonstrated the effectiveness of BBR congestion control in this setup

Abstract

This paper presents a detailed case study of the T2_BR_SPRACE storage frontend architecture and its observed performance in high-intensity data transfers. The architecture is composed of a heterogeneous cluster of XRootD [1] Virtual Machines (VMs) with 10 Gb/s and 40 Gb/s links, which aggregate data from a 77 Gb/s dCache [2] backend via pNFS to an external 100 Gb/s WAN link. We describe the system configuration, including the use of the BBR [3] congestion control algorithm and TCP extensions [4]. Under peak production conditions, we observed the system sustaining an aggregate throughput of 51.3 Gb/s. An analysis of a specific data flow to Fermilab (FNAL) showed peaks of 41.5 Gb/s, validated by external monitoring tools (CERN). This study documents the performance of a complex virtualized architecture under real load.