Where to run boost server?

[monachus]

I'm having some issues with Boost (2.4, 0.8.7e) that are more architectural than anything. I currently run Cacti on a pair of systems clustered with heartbeat and DRBD, which do all the polling, updating, and HTTP serving of the content. I have MySQL running on a different cluster. I'd like to split Cacti up to get better performance, which is what led me to Boost. As I understand it, there's no distributed poller, so the poller has to run on the machine where the RRDs actually live. Presumably, then, the Boost Server has to run here as well?

This is where I'm confused. If I have to run the Boost Server on the poller host, then what am I offloading? Content serving? It looks like the "web farm" referred to in the install manual would simply be connecting to the server and saying "update these RRDs" which is just another layer on top of having the server figure out it should update the RRDs itself.

Since it seems so obviously not helpful, I can only presume I've got it wrong. If someone can shed some light on where the Boost Server would run in this and if it will actually take processing load off of the polling server, that would be tremendous.

Adrian

[TheWitness]

Boost server runs on your poller host typically. It is there only in cases where the Apache user does not have write access to the RRDfiles. In cases where this is true, then the boost server acts as your surrogate making those changes, then subsequently turning control back to you to render the graph using the NFS mount on the non-poller host.

So a load balancing scenario includes an exported files system, mounted on the remote nodes, and simply making requests of the boost server to update the RRDfile and then generating the graph locally using the data from the NFS mount. Again, if you have Write Access on the other Web Servers, its is not required to use the Boost Server.

TheWitness

[monachus]

Hot. I thought that all of this was to somehow avoid the need for NFS mounts and such. It makes more sense now. Thanks!