A prototype Infrastructure for Cloud-based distributed services in High Availability over WAN

21st International Conference on Computing in High Energy and Nuclear Physics

物理学;计算机科学

Bulfon, C.^1 ; Carlino, G.^2 ; De Salvo, A.^1 ; Doria, A.^2 ; Graziosi, C.^1 ; Pardi, S.^2 ; Sanchez, A.^2 ; Carboni, M.^3,4 ; Bolletta, P.^3 ; Puccio, L.^3 ; Capone, V.^5 ; Merola, L.^6

INFN-Roma Unit, Italy^1

INFN-Napoli Unit, Italy^2

GARR the Italian NREN, Italy^3

INFN-LNF Unit, Italy^4

GÉANT, Cambridge, United Kingdom^5

University of Naples Federico II, Italy^6

关键词: Computing infrastructures;    Distributed file systems;    Distributed service;    End-to-end service;    High availability;    Network infrastructure;    Performance evaluations;    Reliability and robustness;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/664/2/022032/pdf
DOI  :  10.1088/1742-6596/664/2/022032

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

In this work we present the architectural and performance studies concerning a prototype of a distributed Tier2 infrastructure for HEP, instantiated between the two Italian sites of INFN-Romal and INFN-Napoli. The network infrastructure is based on a Layer-2 geographical link, provided by the Italian NREN (GARR), directly connecting the two remote LANs of the named sites. By exploiting the possibilities offered by the new distributed file systems, a shared storage area with synchronous copy has been set up. The computing infrastructure, based on an OpenStack facility, is using a set of distributed Hypervisors installed in both sites. The main parameter to be taken into account when managing two remote sites with a single framework is the effect of the latency, due to the distance and the end-to-end service overhead. In order to understand the capabilities and limits of our setup, the impact of latency has been investigated by means of a set of stress tests, including data I/O throughput, metadata access performance evaluation and network occupancy, during the life cycle of a Virtual Machine. A set of resilience tests has also been performed, in order to verify the stability of the system on the event of hardware or software faults. The results of this work show that the reliability and robustness of the chosen architecture are effective enough to build a production system and to provide common services. This prototype can also be extended to multiple sites with small changes of the network topology, thus creating a National Network of Cloud-based distributed services, in HA over WAN.