Error containment cluster of nodes

A multi-processor computer system having a plurality of nodes each having at least one processor and at least one memory device, the system comprising:

• the plurality of nodes connected into a topology having interconnected clusters thereof, the topology isolating first selected nodes from other selected nodes;
• second selected nodes connected from within the topology according to a staggered connection pattern so as to further isolate nodes from each other within the topology;
• means for establishing regions of protected memory and unprotected memory within the clusters;
• and
• means for partitioning memory across clusters so that (1) a particular cluster's protected memory is write-accessible only by nodes within said particular cluster, and (2) said particular cluster's unprotected memory is write-accessible by nodes outside said particular cluster.

A method for containing errors in a multi-processor computer system having a plurality of nodes, each node having at least one processor and at least one memory device, the method comprising the steps of:

• connecting the plurality of nodes into a topology having interconnected clusters thereof, the topology isolating first selected nodes from other selected nodes, the clusters formed by connecting nodes into topologically independent rings thereof, and said interconnection of clusters is established by intersecting the rings at common nodes therein, the topology further established in a Scalable Coherent Interface (SCI) environment;
• connecting second selected nodes from within the topology according to a staggered connection pattern wherein rings are maintained topologically isolated at selected common nodes;
• establishing regions of protected memory and unprotected memory within the clusters;
• and
• partitioning memory across clusters so that (1) a particular cluster's protected memory is write-accessible only by nodes within said particular cluster, and (2) said particular cluster's unprotected memory is write-accessible by nodes outside said particular cluster.