A data caching model for Tier 2 WLCG computing centres using XCache

Teng Li,R Currie ,A Washbrook

doi:10.1051/epjconf/201921404047

Abstract

As the WLCG continues to consolidate the data storage resources to reduce maintaining costs, more storage elements will be decommissioned or federated in the future. In both cases data streaming would become a common method for data accessing in many LHC Tier 2 and Tier 3 computing centres which makes it vital to optimise the network usage (i.e. latency reduction and bandwidth utilisation). Consequently, the data cache service sits closely to the computing resources becomes a desirable tool. In this paper we discuss how to setup a XCache server at an example WLCG Tier 2 site (Edinburgh) and use the ATLAS analysis queue as a testbed to study the data caching performance. The results could be used to guide the future development to make the caching server more effective.

Highlights

MotivationDuring LHC Run 1 and Run 2, the WLCG [1] mostly applied the computing model that colocates the computing and storage resources
There are many available data caching solutions, and in this paper, we focus on the XRootD proxy cache (XCache) [6, 7] as it has the mostly desired
In order to measure the caching performance metrics, an XCache server instance is deployed at an example WLCG Tier 2 site (Edinburgh)

Summary

Motivation

During LHC Run 1 and Run 2, the WLCG [1] mostly applied the computing model that colocates the computing and storage resources. The WLCG is consolidating data storage resources to reduce maintenance costs. As a result of both facts, streaming data option becomes more popular than static data placement, which makes the network the third resource apart from processors and data storage. Optimisations of the network usage become vital, including latency (RTT) reduction and better use of the bandwidth, which usually has huge impacts on the job performance. Data caching is usually a cheap and effective way to optimise the network utilization. We discuss the deployment procedures of XCache, and use the ATLAS [8] analysis queue as a testbed to evaluate the caching performance, which could be referred to optimise the data caching service. The results could be used for other data caching solutions as well

XRootD proxy cache

Deployment

Performance measurement

Findings

Summary