IBM NeXtScale System Planning And Implementation Manual page 174

Install (for nodes with hard disk drives), or push out (for diskless nodes) an
operating system to all servers to verify functionality of the following
components:
– Server hardware
– Ethernet, InfiniBand, and Fibre Channel switches, terminal servers, and
KVM switches
– Server configuration correctness, including memory configurations,
correct CPU types, storage, and RAID
Perform highly parallel linear algebra package (HPLinpack) benchmarking on
the system. This testing assures the following conditions are met:
– CPU and memory are stressed to the extent that is commonly found in
production environments.
– Thermal stress testing is performed.
– Interconnect networks are exercised. HPLinpack is run over the highest
bandwidth, lowest latency network.
As an added benefit, clients can see a performance baseline
measurement, if requested.
Note: HPLinpack is run with open source libraries and no tuning
parameters. Optimization is recommended for those interested in
determining maximum system performance.
This testing is performed with the goal of detecting and correcting all system
issues before the cluster is shipped so that the time from installation to
production is minimized. This testing is meant to verify the hardware and cabling
only. Any software installations require more Cluster Enablement Team, IBM
Global Services, or third-party services.
Note: All servers with local disks have boot sectors wiped before they are
shipped. All shared storage devices have test arrays deconstructed.
156
IBM NeXtScale System Planning and Implementation Guide