Server Reliability Categories and Criteria: Doug Bossen, IBM at IRPS Tutorial 2002 Clarification

1
Server Reliability Categories and CriteriaDoug
Bossen, IBM at IRPS Tutorial 2002Clarification
Pia Sanda, 2/15/05 HPCA (sanda_at_us.ibm.com)

• Undetected errors corrupting customer data
• Typical system target 1000 yr MTBF (114 FITS)
• Detected errors causing system termination
• Corrupt data/control state in a global resource
• Target 25 yr MTBF (4500 FITS unrecovered)
• Detected errors causing application or partition
  termination
• Corrupt data/control state in a local resource
• Target 10 yr MTBF (11400 FITS unrecovered)
• MTBF numbers above are for a HE system scenario,
  not intended to be targets for Power4 or any
  other IBM product.
• Above is for the WHOLE SYSTEM All SYSTEM
  HARDWARE and Operational SOFTWARE (not including
  customer applications)
• HW microprocessors, memory, all other active
  components, interconnects, HDD, power, etc.
• Above is for both HARD and SOFT ERRORS.
• Apportionment of single processor chip cannot be
  inferred from above scenario.
• Transient particle induced contribution is but a
  small part of the scenario targets.