Podcast
Questions and Answers
Static networks utilize switches for message transmission.
Static networks utilize switches for message transmission.
False
Dynamic networks are also referred to as indirect networks.
Dynamic networks are also referred to as indirect networks.
True
Communication latency includes only the transmission time.
Communication latency includes only the transmission time.
False
Bisection bandwidth is calculated by multiplying bisection width by the bandwidth of links.
Bisection bandwidth is calculated by multiplying bisection width by the bandwidth of links.
Signup and view all the answers
In shared-address-space platforms, not all memory is accessible to all processors.
In shared-address-space platforms, not all memory is accessible to all processors.
Signup and view all the answers
Latency in a network refers to the cost of building the network.
Latency in a network refers to the cost of building the network.
Signup and view all the answers
Static networks typically have a higher connectivity than dynamic networks.
Static networks typically have a higher connectivity than dynamic networks.
Signup and view all the answers
Cost in a network is determined by the number of communication links.
Cost in a network is determined by the number of communication links.
Signup and view all the answers
Uniform Memory Access is applicable for real-time applications.
Uniform Memory Access is applicable for real-time applications.
Signup and view all the answers
In Non-uniform Memory Access, memory access time is equal.
In Non-uniform Memory Access, memory access time is equal.
Signup and view all the answers
Message-Passing Platforms consist of a set of processors with shared memory.
Message-Passing Platforms consist of a set of processors with shared memory.
Signup and view all the answers
Message passing requires significant hardware support to operate.
Message passing requires significant hardware support to operate.
Signup and view all the answers
The time to communicate a message in a network includes both message preparation and traversal time.
The time to communicate a message in a network includes both message preparation and traversal time.
Signup and view all the answers
The startup time (ts) only includes the sending of the message.
The startup time (ts) only includes the sending of the message.
Signup and view all the answers
Per-hop time (th) is the time it takes for a message to reach the next node after it has left the current node.
Per-hop time (th) is the time it takes for a message to reach the next node after it has left the current node.
Signup and view all the answers
Shared address space platforms can be implemented efficiently with little effort.
Shared address space platforms can be implemented efficiently with little effort.
Signup and view all the answers
Uniform Memory Access (UMA) is faster than Non-uniform Memory Access (NUMA).
Uniform Memory Access (UMA) is faster than Non-uniform Memory Access (NUMA).
Signup and view all the answers
In a Uniform Memory Access system, a single memory controller is utilized.
In a Uniform Memory Access system, a single memory controller is utilized.
Signup and view all the answers
Non-uniform Memory Access (NUMA) achieves a limited bandwidth compared to UMA.
Non-uniform Memory Access (NUMA) achieves a limited bandwidth compared to UMA.
Signup and view all the answers
A system classified as NUMA will have the same access time for all memory locations.
A system classified as NUMA will have the same access time for all memory locations.
Signup and view all the answers
Cache-only Memory Access (COMA) is one of the three shared memory models categorized alongside UMA and NUMA.
Cache-only Memory Access (COMA) is one of the three shared memory models categorized alongside UMA and NUMA.
Signup and view all the answers
Uniform Memory Access systems provide a uniform time to access any memory in the network.
Uniform Memory Access systems provide a uniform time to access any memory in the network.
Signup and view all the answers
In NUMA, memory access time varies due to the use of multiple memory controllers.
In NUMA, memory access time varies due to the use of multiple memory controllers.
Signup and view all the answers
All shared-address-space architectures utilize a uniform memory access protocol.
All shared-address-space architectures utilize a uniform memory access protocol.
Signup and view all the answers
The per-hop time is defined as the time taken by the header of a message to travel between two directly-connected nodes in a network.
The per-hop time is defined as the time taken by the header of a message to travel between two directly-connected nodes in a network.
Signup and view all the answers
The per-word transfer time is calculated as $tw = r$.
The per-word transfer time is calculated as $tw = r$.
Signup and view all the answers
For a message of size m words traversing l communication links, the total communication cost includes both the header cost and the cost of the message transfer.
For a message of size m words traversing l communication links, the total communication cost includes both the header cost and the cost of the message transfer.
Signup and view all the answers
In parallel computers, the per-hop time is generally larger than the per-word transfer time for small values of m.
In parallel computers, the per-hop time is generally larger than the per-word transfer time for small values of m.
Signup and view all the answers
A Parallel Random Access Machine (PRAM) consists of a single processor and a limited global memory.
A Parallel Random Access Machine (PRAM) consists of a single processor and a limited global memory.
Signup and view all the answers
Interconnection networks in parallel computers are solely made of processors.
Interconnection networks in parallel computers are solely made of processors.
Signup and view all the answers
Processors in a PRAM operate independently and do not execute different instructions at each cycle.
Processors in a PRAM operate independently and do not execute different instructions at each cycle.
Signup and view all the answers
The cost incurred by the header for a message traversing multiple links is typically negligible in current parallel computing scenarios.
The cost incurred by the header for a message traversing multiple links is typically negligible in current parallel computing scenarios.
Signup and view all the answers
Study Notes
Interconnection Networks
- Two main types of interconnection networks: Static Networks and Dynamic Networks.
- Static Networks consist of point-to-point communication links; also known as Direct Networks.
- Dynamic Networks involve switches and allow for dynamic connection of communication links; also referred to as Indirect Networks.
Network Criteria
- Latency includes transmission time and overhead related to communication.
- Bandwidth defines the number of bits transmitted per second.
- Connectivity measures the number of paths between any two processors.
- Cost is determined by the number of links in the network.
- Ease of Construction relates to the network's regularity.
Comparing Static Networks
- Diameter indicates the maximum shortest path between any two processors, reflecting worst-case latency.
- Bisection Width points out the minimum number of links that must be severed to split the network into two equal parts, highlighting potential bottlenecks.
- Bisection Bandwidth is calculated as the product of bisection width and bandwidth of those links.
Communication Model of Parallel Platforms
- Data exchange methods include accessing a shared data space and message passing.
- Shared-address-space machines allow all processors to access the same memory; they can be classified as either UMA or NUMA.
- UMA (Uniform Memory Access) uses a single memory controller and provides balanced access time, though it is slower than NUMA.
- NUMA (Non-Uniform Memory Access) utilizes multiple memory controllers, is generally faster, and better suited for high-performance applications.
Shared-Address-Space Architectures
- Three types of shared memory models: UMA, NUMA, and Cache-only Memory Access (COMA).
- In UMA, all memory access times are equal; used for general-purpose applications.
- In NUMA, memory access times can vary; suited for applications requiring high bandwidth.
Message-Passing Platforms
- Comprise multiple processors with exclusive memory, often employed in clustered workstations (non-shared address-space).
- Utilize send and receive message variants for communication, requiring minimal hardware support compared to shared-address-space platforms.
Communication Costs in Parallel Machines
- Communicating a message includes preparation time and transmission time.
- Startup time encompasses the handling time at both sending and receiving nodes.
- Per-hop time refers to the time taken for a message to travel from one node to the next.
- Per-word transfer time is influenced by channel bandwidth, representing the duration needed for each word to traverse a link.
Costs of Message Transmission
- The total communication cost combines per-hop cost and per-word transfer costs for multiple links.
- The equation for total communication time involves startup time and transfer time across the number of links.
Physical Organization of Parallel Platforms
- PRAM (Parallel Random Access Machine) is an abstract model allowing synchronous operation with shared memory access.
- In PRAMs, processors work under a common clock, executing potentially different instructions each cycle.
Interconnection Networks for Parallel Computers
- Essential for data transport between processors and memory.
- Composed of switches and links, which can be wired or fiber-based, facilitating high-speed communication.
Studying That Suits You
Use AI to generate personalized quizzes and flashcards to suit your learning preferences.
Related Documents
Description
Test your understanding of message passing architectures, focusing on static and dynamic interconnection networks. This quiz covers the criteria of network performance, including latency and various network configurations. Assess your knowledge on direct and indirect networks!
