CS 470- Multiprocessing
- Explicit vs. implicit
- Threads vs. processes
- Thread safety
- Non-determinism
- Race conditions
- Deadlock
- Synchronization
- Mutual exclusion
- Locks
- Semaphores
- Conditions
- Monitors
- Barriers
- OpenMP
- Pragmas
- parallel
- for
- task
- single
- master
- critical
- barrier
- Clauses
- default/private/shared
- reduction
- schedule
- collapse
- firstprivate/lastprivate
- nowait
- Functions
- omp_get_wtime
- omp_get_num_threads
- omp_get_max_threads
- Loop-carried dependencies
- Teams and parallel regions
- Locks
- Parallel languages
- Productivity vs. performance
- Partitioned global address spaces (PGAS)
- High-Performance Fortran
- CAF and UPC
- X10 and Fortress
- Chapel
- Python and Julia
- Performance analysis
- Speedup and efficiency
- Amdahl's and Gustafson's laws
- Linear speedup
- Critical path analysis
- Temporal vs. spatial locality
- Weak vs. strong scaling
- CPU time vs. wall time
- Communication overhead
- Bandwidth
- Latency
- Bisection
- Contention
- Energy usage
- Energy (work) vs. power (rate)
- Power caps
- Dynamic voltage frequency scaling
- Analysis tools
- Debuggers
- Profilers
- Hardware counters
- Timer resolution
- Sampling
- Overhead
- Perturbation
- Skid
- Tool frameworks
- Performance modeling
- Roofline model
- Autotuning
- Distributed issues
- Naming
- Flat namespaces
- Hierarchical namespaces
- IPv4, IPv6, and DNS
- Partitioned global address spaces (PGAS)
- Overlay networks
- Distributed hash tables
- Virtual address space
- Finger / lookup tables
- Chord
- Synchronization
- Message passing
- Clocks
- Physical
- Lamport clocks
- Vector clocks
- NTP
- Barriers
- Consensus protocols
- Transactions
- Elections
- One-phase vs. two-phase
- Paxos
- Replication and consistency
- Partial vs. total orderings
- Data-centric
- Continuous
- Sequential
- Causal
- Client-centric
- Monotonic reads
- Monotonic writes
- Read-your-writes
- Writes-follow-reads
- Distributed version control
- Fault tolerance
- CAP theorem
- Consistency
- Strong
- Eventual
- Weak
- Availability
- Active / passive
- Active / active
- Partition tolerance
- Soft vs. hard failure
- Permanent vs. intermittent vs. transient faults
- MTBF and FIT
- Failure types
- Crash
- Omission
- Timing
- Response
- Arbitrary (Byzantine)
- Failure handling
- Detection
- Prevention
- Avoidance
- Recovery
- Techniques
- DMR vs. TMR
- Checksums / hashes
- Hamming codes
- Reed-Solomon codes
- Checkpointing
- Security
- Attacks
- Brute force password cracking
- Replay attacks
- Man-in-the-middle attacks
- Principle of least privilege
- Trust
- Policies
- Encryption
- One-way hash functions
- Cryptographic systems
- Symmetric vs. asymmetric
- MD5 / SHA
- DES / RSA
- Authentication
- Shared-key challenge/response
- Needham-Schroeder
- Kerberos
- Key exchange parties
- Diffie-Helman key exchange
- Certificate authorities
- Authorization
- Firewalls
- Unix file permissions
- Access control lists
- LDAP and AD
- Auditing
- Append-only logs
- Blockchains (Bitcoin)
- File systems
- Design issues
- File-level vs. block-level
- Remote access vs. upload/download
- Centralized vs. decentralized
- Symmetric vs. asymmetric
- Striping
- Remote procedure calls
- Function stubs
- Parameter marshalling
- Synchronous vs. asynchronous
- Networked file systems
- Exports
- Mounts
- Static vs. automatic
- Protocols
- NFS
- AFS
- GoogleFS
- Lustre
- Bittorrent
- Freenet
- Middleware
- Scheduling
- SLURM
- Interactive vs. batch jobs
- Parameterized MPI jobs
- Monitoring
- Load balancing
- Checkpoint/restart
- Parallel patterns and concepts
- Task vs. data decomposition
- Shared-memory vs. distributed-memory
- Locality
- Data access patterns
- Spatial vs. temporal locality
- NUMA effects
- Caching
- Mirroring
- Content delivery networks
- Foster's methodology
- Partitioning
- Communication
- Aggregation
- Mapping
- Communication patterns
- Naturally ("embarrassingly") parallel
- Reduction trees
- Nearest-neighbor
- Producer/consumer
- Map/reduce
- Pipelines and streams
- Collective operations
- Broadcast
- Reduction
- Scatter
- Gather
- Allgather
- Allreduce
- All-to-all
- Matrix operations
- Sparse vs. dense
- Access patterns
- Linear system solvers
- Linear algebra
- Architectures and technologies
- Flynn's taxonomy
- SISD
- SIMD
- MIMD
- SPMD
- SIMT
- Instruction-level parallelism
- von Neumann bottleneck
- Pipelining instructions
- Superscalar processing
- Speculative execution
- Vector processing
- Shared memory
- Threading libraries
- Pthreads
- Java threads
- Windows threads
- OpenMP
- Manycore
- Coprocessors and accelerators
- GPUs / GPGPUs
- SIMT
- Streaming multiprocessors
- Warps and divergence
- Host vs. device memory
- CUDA
- Kernels
- Thread blocks and grids
- Grid-stride access pattern
- Atomic operations
- Fast barrier
- OpenACC
- Distributed clusters
- OpenMPI and MPICH
- Homogeneous vs. heterogeneous
- Hybrid w/ accelerators
- Topologies
- Bus
- Crossbar switches
- Star
- Ring
- Grid / Mesh
- Torus
- Hypercube
- Fat trees
- Interconnects
- Ethernet
- InfiniBand
- OmniPath
- Supercomputers
- Wide-area networks
- End-to-end principle
- Sockets
- OSI model
- QoS concerns
- Routing
- Circuit switching vs. packet switching
- Unicast
- Multicast
- Broadcast
- Web protocols
- IP / DNS
- TCP / UDP
- HTTP / HTML
- SSL / TLS
- NTP
- XML / SOAP / JSON
- Peer-to-peer
- Bittorrent
- Tor
- Freenet
- Clouds and grids
- Infrastructure-as-a-service
- Virtualization
- Type-1 vs. type-2 hypervisors
- Virtual machines
- Containers / Docker
- Cloud providers
- Amazon AWS
- Google Cloud
- Microsoft Azure
- Rackspace
- Work-sharing
- Condor
- GIMPS
- Novel architectures
- Chiplets
- Memory-centric
- Neuromorphic
- Quantum
- Optical
- Nanosheet transistors
- Turnkey AI solutions