Fault Tolerant Multiprocessor and Distributed Systems

1
Fault Tolerant Multiprocessor and Distributed
Systems
2
Fault Tolerant Multiprocessor and Distributed
Systems

• Dependability is designed in a hierarchical
  fashion
• Circuit and technology levels
• Node-level architecture
• Internode architecture
• Operating system level
• Applications level

3
Classification of Multiprocessors

• Shared memory versus distributed memory
• Sequent Symmetry, Encore Multimax, Sequoia,
  Stratus shared memory
• Intel iPSC hypercube, Intel Paragon, CM-5, IBM
  SP-1, Tandem Nonstop distributed memory
• Fine grain versus coarse grain
• CM-2 synchronizes after every clock cycle fine
  grained
• Intel iPSC synchronizes after several hundred to
  thousand clock cycles medium grained
• Distributed network of workstations coarse
  grained

4
Classification of Multiprocessors
5
Classification of Multiprocessors

• Fine grain versus coarse grain
• CM-2 synchronizes after every clock cycle fine
  grained
• Intel iPSC synchronizes after several hundred to
  thousand clock cycles medium grained
• Distributed network of workstations coarse
  grained

6
Classification of Multiprocessors

• SIMD versus MIMD
• CM-2, Masspar MP-1 SIMD
• Intel iPSC, CM-5, BBN Butterfly MIMD
• Topology of interconnect
• Typical interconnection networks bus, crossbar,
  multistage networks, rings, meshes, tress, and
  hypercubes
• Programming model

7
Fault Tolerance in Multiprocessors

• Static or Masking redundancy
• Dynamic or standby redundancy

8
Static Redundancy

• Three distinct applications of static redundancy
  in multiprocessors
• Redundancy for reliability and availability
• Redundancy for safety
• Redundancy for tolerating non-classical faults

9
Redundancy for Reliability and Availability

• Each processor can be replicated
• Entire multiprocessor can be replicated
• Divide P processors into P/N groups of N
  processors, each group voting on its results
  before communicating to other groups
• For robust communication critical transactions
  between groups may be replicated and voted

10
Redundancy for Safety

• Reliability-safety trade-off

11
Redundancy for Safety

• Safe Modular Redundancy (SMR)
• SMR with n modules - nSMR