1/95
Looks like no tags are added yet.
High-Performance Computing (HPC)
What is the process of aggregating computing power to solve problems that are too large, complex, or time-critical for a workstation?
Capability and Capacity
What two objectives does HPC achieve by aggregating computing power?
Computational Complexity
What measures how the time and resources required by an algorithm grow as the problem size increases?
Big O Notation
What notation is usually used to express time complexity?
O(n³)
What computational complexity represents cubic growth, where work increases with the cube of the problem size?
O(2ⁿ)
What computational complexity represents exponential growth, where work doubles with each additional input?
Computational Complexity
What helps estimate computation time, memory requirements, and the potential benefits of parallelization?
Data Scale
What refers to the volume of data that must be generated, stored, transferred, and analyzed during computation?
Lustre, GPFS (Spectrum Scale), and BeeGFS
What are the three parallel file systems used in HPC for storage?
Data Movement Bottleneck
What occurs when transferring data takes longer and costs more than computing on it?
Machine Learning and Data Analytics
What techniques help scientists identify patterns, detect anomalies, and extract insights automatically from massive datasets?
Time Constraint
What is the maximum acceptable time to complete a computation and deliver a useful result?
Time-to-Solution
What HPC metric measures how long it takes to obtain the final answer for a given problem?
FLOPS
What metric measures floating-point operations per second?
Compute Nodes
What are the main building blocks of an HPC system?
CPU, GPU/Accelerators, and Main Memory (RAM)
What are the three main components found inside a compute node?
Rack
What houses multiple compute nodes and provides organized deployment, shared power and cooling, and easier maintenance?
Cluster
What is a collection of racks working together as one HPC system?
Interconnect
What communication fabric connects all nodes and racks in an HPC system?
MPI Message Passing, Node-to-Node Communication, and Fast Data Transfer
What are the responsibilities of an HPC interconnect?
Storage
What HPC component provides shared access to data across the entire cluster?
Node
Which HPC component performs computations?
Rack
Which HPC component groups nodes?
Cluster
Which HPC component groups racks?
Interconnect
Which HPC component enables communication between nodes and racks?
Storage
Which HPC component holds and serves data?
Lustre
What is the high-performance parallel distributed file system designed for large-scale HPC clusters and supercomputers?
Lustre
What file system allows thousands of compute nodes to access the same data simultaneously at high speeds?
Metadata Target (MDT)
What Lustre component stores file system metadata such as filenames, directories, and permissions?
Object Storage Targets (OSTs)
What Lustre components store actual file data across multiple storage targets?
LNET (Lustre Networking)
What provides the network connection for Lustre clients and enables high-performance parallel I/O?
Task Parallelism
What type of parallelism involves executing different tasks concurrently, such as atmosphere, ocean, and land climate models?
Data Parallelism
What type of parallelism performs the same operation on different sets of data?
Amdahl's Law
What law predicts the maximum speedup achievable by parallelizing a program?
S = 1 / ((1 − P) + P/N)
What is the formula for Amdahl's Law?
P
In Amdahl's Law, what represents the parallel fraction of a program?
N
In Amdahl's Law, what represents the number of processors or cores?
10× Speedup
What is the maximum speedup when 10% of a program must execute sequentially, even with unlimited processors?
Gustafson's Law
Assumes that when more processors become available, scientists solve larger problems instead of keeping the workload fixed
Amdahl's Law
Assumes the workload remains fixed. If part of a program is sequential, it limits the maximum speedup.
Scalability
What concept is emphasized by Gustafson's Law?
Speed Limitation
What concept is emphasized by Amdahl's Law?
MPI (Message Passing Interface)
What is the standard programming model used in HPC for communication between processes running on different nodes?
InfiniBand
What network technology can be used for communication between nodes in an HPC cluster?
Send Messages, Receive Messages, Synchronize Processes, and Share Results
What are the four functions MPI provides for communication between nodes?
OpenMP (Open Multi-Processing)
What API is used for shared-memory parallel programming on multicore CPUs?
Threads
What does OpenMP create to allow a program to use multiple CPU cores within one node?
CUDA (Compute Unified Device Architecture)
What is NVIDIA's parallel computing platform and programming model for general-purpose GPU computing?
Host (CPU)
In CUDA's execution model, what component controls the program?
Device (GPU)
In CUDA's execution model, what component performs the heavy computations?
Hybrid Parallel Programming Model
What programming model combines multiple parallel programming technologies to utilize modern HPC systems?
MPI + OpenMP + CUDA
What three technologies are commonly combined in a hybrid HPC programming model?
MPI
Which hybrid programming technology handles distributed memory, message passing, and cluster-level parallelism?
OpenMP
Which hybrid programming technology handles shared-memory parallelism and multiple CPU cores within a node?
CUDA
Which hybrid programming technology accelerates compute-intensive tasks such as AI, simulations, and matrix operations?
Communication Overhead
What is the time spent sending messages, receiving messages, and synchronizing data instead of computing?
Load Imbalance
What occurs when work is not distributed equally among processors, causing some processors to finish early and wait?
Slowest Processor
Which processor determines when the entire parallel job finishes when workloads are imbalanced?
Memory Wall
What occurs when computation speed grows faster than memory speed, causing CPUs to wait for data?
Overhead
What refers to time spent communicating, waiting, or fetching data instead of computing?
Weak Scaling
What type of scaling measures how well a system handles a larger problem as more processors are added while keeping the execution time approximately constant?
Fixed Problem Size
What remains constant in strong scaling?
Growing Problem Size
What increases with the number of processors in weak scaling?
PAGASA
What organization is used in the file as an example of an application that needs strong scaling for time-critical weather forecasting?
MPI, OpenMP, and CUDA
What parallel programming technologies are combined in the climate simulation example?
Speedup
What performance measure identifies how many times faster an HPC system completes the same task?
Execution Time
What performance measure determines how much the processing time decreases?
Efficiency
What performance measure identifies how effectively available CPU and GPU resources are utilized?
Throughput
What performance measure identifies how many tasks or calculations can be completed within a given period?
Scalability
What performance measure determines how well performance improves when additional processors or computing resources are added?
DOST-ASTI COARE
What Philippine HPC facility is presented in the architecture diagram as the Computing and Archiving Research Environment?
Computing and Archiving Research Environment
What does COARE stand for?
44 Nodes
What is the total number of nodes in the COARE architecture diagram?
36 CPU Nodes and 8 GPU Nodes
What is the breakdown of the 44 COARE nodes?
3,168 CPU Cores
How many total CPU cores are indicated for the COARE CPU nodes?
88 Logical CPUs per Node
How many logical CPUs does each COARE CPU node have?
24 CPUs per Node
How many CPUs are indicated for each P40 GPU node?
1 TB RAM per Node
How much RAM is indicated for each P40 GPU node?
NVIDIA Tesla P40
What GPU model is installed in the six P40 nodes?
6 P40 Nodes
How many P40 nodes are indicated in the COARE architecture diagram?
NVIDIA Tesla A100
What GPU model is installed in the two A100 nodes?
8 GPUs per Node
How many NVIDIA Tesla A100 GPUs are indicated per A100 node?
128 CPUs per Node
How many CPUs are indicated for each A100 node?
1 TB RAM per Node
How much RAM is indicated for each A100 node?
Approximately 75 TFLOPS
What is the combined GPU compute capacity indicated for the A100 nodes?
Approximately 30 TFLOPS
What compute performance is indicated for the P40 nodes?
10 Gb/s Ethernet
What network speed is indicated for the COARE management and storage network?
SLURM
What workload manager is used for resource management in the COARE cluster?
InfiniBand 200 Gb/s
What interconnect technology and link speed are indicated in the COARE architecture?
1–3 Microseconds
What is the indicated low latency of the COARE InfiniBand interconnect?
Lustre Parallel File System
What storage system is used by COARE to provide high-throughput parallel data access?
Approximately 1,015 TB
What is the total Lustre storage capacity indicated in the COARE architecture diagram?
64–512 GB RAM
What is the general RAM range indicated for compute nodes in the educational HPC architecture overview?
Local NVMe SSD
What type of fast local storage is included in the compute node architecture?
Parallel File System
What type of storage system enables multiple compute nodes to access shared data simultaneously?
AI, Weather Modeling, and Data Science
What three research areas are identified as applications supported by DOST-ASTI COA