|Related to : How to make a custom server run in a cluster|
|How to set up automatic failover to a standby MT cluster from a dedicated cluster?|
There are a few options
Install a load balancer (e.g. Amazon ELB) and connect the two
Cloudant cluster A (the primary cluster)
Cloudant cluster B (the MT backup cluster)
The load balancer can then load balance between the two, or do logic
like "if I can't see B, send traffic to A".
Build functionality to handle switchover in your application layer but
arguably using a LB is cleaner
|Custom Server with peerjs - no documentation has left me clueless|
You don't handle calls in the server.js file. The peer server is only
for signalling. See this basic signalling server
You just need this to connect browser peers and make video connections
Try the examples on peerjs.com and work your way up, ie:
make a call:
var getUserMedia = navigator.getUserMedia ||
navigator.webkitGetUserMedia || navigator.mozGetUserMedia;
|Zookeeper cluster set up|
have you started zookeeper in all the three nodes ? In a multi-cluster
set up (assuming you have a distributed environment with multiple
machines) every server knows about the other nodes present in the
cluster known as ensemble. It does this by looking at the following
piece of line in the zoo.cfg file.
In multi-cluster set up doc page it
|How to find the right cluster algorithm?|
Try Mclust from the mclust package, it will try to fit a Gaussian
mixture on your data.
The default behavior:
mc = Mclust(matrix);
.. will find 4 groups, but you might be able to force it to 2 or to
force the correlation structure for the Gaussians to be stretched in
the right direction.
Be aware that it'll be hard to interpret and justify the meaning o
|Cluster variables together and run ANOVA|
I don't have the code on me, but originally I was aiming for a
sophisticated method to cluster variables together with minimal
errors. Instead, I opted for two methods when combining the variables:
1) averaging the groups; 2) summing up the groups. One should note,
however, that the selected method has an impact on the output (i.e.,
beta weights, marginal means, etc.) - p-values remain the same.