search

duclos-cavalcanti / TreeBuilder

Master Thesis on High Performance Multicasting on the Cloud
BSD 3-Clause "New" or "Revised" License
1 stars 0 forks source link

K Parameter - Pool Slicing #8

Open duclos-cavalcanti opened 4 months ago

duclos-cavalcanti commented 4 months ago

Discuss if necessary, as to bound how much children a parent probes, limiting stress placed on NIC.

nrasadi commented 1 month ago

@duclos-cavalcanti
I am unsure what this issue is about 🤔
Have we discussed this?

duclos-cavalcanti commented 1 month ago

@duclos-cavalcanti I am unsure what this issue is about 🤔 Have we discussed this?

Sorry for the issue not being very clear.

This is basically referring to the K parameter, which specific how many nodes from the Pool will the tool (Manager) take out to perform an examination of possible children to a node (Parent Job).

Currently K is fixed to be 2 times F (Fanout) of the tree which effectively means the manager tool will always probe 2x the amount of possible children a node can have and then select the best half of those nodes.

I just placed this issue for future discussions if the way K is defined makes sense or not. The alternative would be that the manager basically probes ALL nodes in a pool and selects the best F (Fanout) nodes. But in a pool of 1000 nodes, that would be cumbersome and not ideal for TreeFinder's lightweight goals.