[gomd-devel] <.PLAN> request for comments... ;)

[Gian Paolo Ghilardi]

Hi all.

These are some considerations about the two remianing features for gomd.

1) I've just started working on the ClusterSnapshot feature.

My idea is to provide a complete status summary for all nodes of the
cluster.
- So each node has its own C++ STL map with all the stats.
- Done this step, gomd will create an HTML (or XML) file to be published
directly via a webserver (i.e. Apache).
- ClusterSnapshot will be executed in a separate thread and the user can
start/stop this fetaure as needed.
- The refresh time of the thread (and the refresh time of the produced
HTML/XML file) can be modified at runtime by the user.

2)I'm thinking about the last planned feature (that is the most difficult
one): BlackHole Detection => autodetecting overloaded/malfunctioning nodes.

We've two options.
- The "standard" alarm-if-threshold-is-passed mechanism (using
ClusterSnapshot to get the values and a config file to specify thresholds).
- Something like using Neural Networks (I was thinking about perceptrons,
naive Bayes, backpropagation,...) .

Comments?

PS: thansk Matt for your help!!! ;)

<rejected>