Query execution and optimization with autonomic error recovery from network failures in a parallel computer system with multiple networks

An article of manufacture for executing on a parallel computer system with a plurality of compute nodes connected by a plurality of independent networks comprising:

• a query execution monitor performing the steps of:
• receiving a query to a database;
• optimizing the query, wherein the query optimizer optimizes the query by generating an execution plan that includes using a network of the plurality of independent networks to execute the query;
• initiating the query execution;
• monitoring the query execution;
• monitoring a plurality of independent networks and recording information about network utilization in a network file;
• detecting a network failure based on the information in the network file indicating a network resource is below a threshold;
• in response to the detected network failure, re-optimizing the query by generating an alternate execution plan for the query to use a different independent network to execute the query where the selection of the different independent network by the query optimizer is based on the information in the network file;
• wherein the network file contains network file information comprising:
• network ID, a timestamp;
• current utilization, future utilization, availability, latency, and retransmits;
• and
• non-transitory computer recordable media in which computer instructions are stored, which instructions, when read by a computer, cause the computer to perform the steps of the query execution monitor.

A computer apparatus comprising:

• a massively parallel computer system with a plurality of nodes each having a memory and at least one processor;
• a plurality of independent networks connecting the plurality of nodes;
• a database residing in the memory;
• a network monitor that periodically monitors the plurality of independent networks to determine network loading and maintains a network file that contains information about network utilization for the plurality of independent networks;
• a query optimizer and a query to the database residing in the memory, wherein the query optimizer optimizes the query by generating an execution plan that includes using a network of the plurality of networks to execute the query;
• a query execution monitor residing in the memory and executed by the at least one processor, the query execution monitor detects poor performance of the network during execution of the query and invokes the query optimizer to re-optimize the query by generating an alternate execution plan for the query to use a different network to execute the query where the selection of the different network by the query optimizer is based on the information in the network file;
• wherein the network file is used by the query execution monitor and wherein the network file contains network file information comprising:
• network ID, a timestamp;
• current utilization, future utilization, availability, latency, and retransmits.