Am Dienstag, 22. Oktober 2013, 22:21:02 schrieb C. Handel: > place a loadbalancer in front of the cluster. Buy hardware, for example F5 > or coyote. Or go for software with haproxy and keepalived. Use ldirectord and pacemaker. Trust me. > The software > loadbalancer is a shared nothing. You only need to keep the config in sync > (for example using csync2). Configure load balancing with sticky sessions > based on incoming ips. You might want to sync the stored connections. ipvs has a demon, that takes care of this. > Create a redhat cluster with four nodes. Main service provided by rhcs is > GFS2. This allows you to share configfiles and web delivery files the same > on all nodes. Not needed to set up a full blown RH cluster. Keep the config files in sync with a simple rsync or csync. No further shared storage needed. Just use corosync for the cluster communication and pacemaker for the resource management. Resources are the ldirectord and the virtual IP. > Start an apache on each node (via clones). Or start it separate on every node via init or systenmd. no cluster needed in the backend web servers. > Traffic goes through loadbalancer and hits all rhcs nodes. Bandwith limit > is the interface of the loadbalancer. Processing limit is from each > RHCS-node. Disk I/O limit is from the SAN backing the GFS2 in combination > with the GFS2 (reading should be no issue, heavy writing will trash the > cluster via locks) Please think twice of you really need a cluster on the backend web servers. In most of the designs you can live without. Which makes live much easier. -- Dr. Michael Schwartzkopff Guardinistr. 63 81375 München Tel: (0162) 1650044 Fax: (089) 620 304 13 -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
