Paralellize calls to components
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
lexiko |
New
|
Medium
|
Nicolas Slusarenko |
Bug Description
- When a call to component is performed it is done something like this:
matrixOpeMask mo= create<
c= mo.product(a, b);
- If there are two 'matrixOpeMask' we could try:
moA.
moB.
c1= moA.productA();
c2= moB.productA();
- We trigger a job (the product of matrices 'a' and 'b') on 'moA' and then we trigger another job on 'moB'.
- However, if conveying the arguments 'a' and 'b' to the component takes a lot of time it would be a great advantage to have an internal thread on the trigger call 'T_product(..)' sending the arguments. Then the call will return immediately, the thread will continue doing the job.
- Then would be possible to do the same with 'moB'.
- This case happened to be relevant sending two matrices, 'a' and 'b' of 1000x1000 (one million of elements) each one.
- Sending the arguments take between 0.5 and 1.0 second and it is not consuming the whole bandwidth, so it is perfectly feasible to have two, or more, calls in parallel.
description: | updated |
description: | updated |
Changed in lexiko: | |
assignee: | nobody → Nicolas Slusarenko (nicolas-slusarenko) |
description: | updated |
Changed in lexiko: | |
importance: | Undecided → Medium |