This project is aimed at reorganising the squid internals to provide a more flexible, intuitive and much faster IO model.
After a good 18 months of playing around with various optimisation ideas, I've settled on this plan:
Once this is done, I'll re-evaluate the client<->server data flow model, since I would like to push the multiple-storeclient code into the storage FS layer and have the squid storage manager deal with a single client to server mapping. This will drastically reduce the amount of code required for the server/client flow, and hopefully help along the implementation of an improved network IO layer.