Design Patterns for Resiliency - The Practice of Cloud System Administration - page 132

Power Supplies
Each machine has a power supply that converts standard electric voltages into levels
needed by the computer. Power supplies frequently die. Servers, network equipment, and
manyothersystemscanbepurchasedwithredundantpowersupplies. N +1and N +2con-
figurations are commonly available.
As with RAID, a strategic use of redundant power supplies is best. They are not needed
when the system itself is a replica or some other resilience technique is used at a higher
level. Do use such power supplies for the remaining systems that are not redundant.
Network Interfaces
Network interfaces or network connections themselves often fail. Multiple links can be
used in N + 1 configurations. There are many standards, too many to detail here.
Some are load sharing, others are active-passive. Some require that all the near-end
(machine)connections bepluggedintothesamenetworkinterface controller(NIC)daugh-
terboard. If two physical ports share the same daughterboard, the failure of one may cause
the other to fail. Some require that all the far-end (switch) connections be plugged into the
same switch, while others do not have such a limit. The latter approach provides resiliency
against switch failure, not just NIC failure.
Many different algorithms are available for determining which packets go over which
physicallink.Withsome,itispossibleforpacketstoarriveoutoforder.Whileallprotocols
should handle this situation, many do not do it well.