Knowledgebase
Knowledgebase
Portal Home > Knowledgebase > Articles Database > Scalable Topology
Scalable Topology
Posted by mikeanthonywild, 11-07-2011, 04:55 PM I am in the process of considering designs for an ultra-dense rackmount server at the moment and have a few questions regarding the topology. I shall endeavour to sum up my idea more clearly: -With a dedicated package a client is entitled exclusive use over a particular quantity of hardware. -The server I have devised contains a huge cluster. -It is not advisable to run Xen on the nodes. -Instead, run Xen on a separate more powerful "head" machine. -If the customer orders 2 CPU units, then two nodes running Lighttpd have the IP of the respective Xen client whitelisted. Similarly the Lighttpd instance running in the Xen client has load balancing set to use the two specified nodes. The aim is to provide something of a cross between a Dedicated and Cloud hosting package. Most requests will be forwarded by the load balancer to a huge cluster of nodes running Lighttpd and the usual web suspects (PHP, Ruby, Python etc.). Due to the nature of the nodes it is not practical to install Xen on each one, so instead I was going to carry out virtualization either on, or just behind the load balancer. The problem that I have is how can I limit the traffic between the Xen client and the cluster nodes? My first thought was to assign each node exclusively to a Xen client, thus being something between a VPS and a dedicated package. The only problem is that after X amount of customers, all the nodes in a cluster would be assigned, but it is very unlikely that they'll all be being used 100% making it rather inefficient. Although I guess the end-user pays a premium for the exclusiveness of the hardware. This is all theoretical really as I have never ventured into this area and so it would be foolish to dive head-first without doing some thorough research first. Last edited by mikeanthonywild; 11-07-2011 at 05:05 PM.
Add to Favourites 
Print this Article
Also Read
