A communication fabric in a parallel computer has the structure of a binary tree. Processors are connected to the leaf nodes and memory system lock hardware is connected to the root. Each internal node in the fabric has the ability to compare simultaneous lock requests from its two child nodes for equality. If the requests are for the same lock and for read-access to a memory location, the fabric node will combine the two requests into one unified request to pass up the tree and remember the combined access. When the memory system responds to the read access, a fabric internal node will, in response to a match with a remembered combined request, replicate the response to “un-combine” and pass the response to both of its child nodes. How does this Combining Tree communication fabric improve parallelism for read-access to lock-protected values in memory
A communication fabric in a parallel computer has the structure of a binary tree. Processors are connected to the leaf nodes and memory system lock hardware is connected to the root. Each internal node in the fabric has the ability to compare simultaneous lock requests from its two child nodes for equality. If the requests are for the same lock and for read-access to a memory location, the fabric node will combine the two requests into one unified request to pass up the tree and remember the combined access. When the memory system responds to the read access, a fabric internal node will, in response to a match with a remembered combined request, replicate the response to “un-combine” and pass the response to both of its child nodes.
How does this Combining Tree communication fabric improve parallelism for read-access to lock-protected values in memory?
Trending now
This is a popular solution!
Step by step
Solved in 2 steps