|
|
The upcoming LHC system at the particle research center CERN will have clusters of O(1000) nodes for each of its four experiments ALICE, ATLAS, CMS and LHCb.
The clusters will run Linux as their OS and the cluster nodes will be connected via a high speed interconnect, such as GbE, Infiniband or Myrinet.
The total disk space capacity of these systems will be in the range of O(1PB) per cluster. The question is how to use this huge disk space in an efficient
way.
The ClusterRAID project presented on this website aims to make the cluster online storage capacity available to applications. Major obstacles are the
diversity of application requirements and the inherent unreliability of individual cluster components (in particular of the hard drives).
The key paradigm of ClusterRAID is, therefore, to make the local hard drive reliable, while preserving the easy-to-use block-level interface of the
underlying hard drive. Key features of this novel distributed architecture are the adjustable degree of fault-tolerance and the minimization of both
network transactions and required space overhead.
|
Search
Print view
Deutsch
University