On May 14, 2007, at 10:25 AM, Ruslan Sivak wrote: > Steve Huff wrote: > > If you set up a third box to be the shared storage, doesn't that > now become the single point of failure? Short answer: maybe. :) Longer answer: If you set up your shared storage according to upstream's guidelines, as described in the documentation (http:// mirror.centos.org/centos/4/docs/html/rh-cs-en-4/ch-hardware.html#TB- HARDWARE-NOSPOF), then you provide at least two channels of communication between each component in the cluster. In addition, you choose a platform for shared storage that provides some redundancy of its own, whether it's multi-controller HW RAID, or multiple storage nodes on a SAN, or what have you. CS/GFS operates under the assumption that your shared storage is fault-tolerant; its job is to make your services fault-tolerant. Is the recommended "no single point of failure" configuration proof against your data center burning down, or against a madman with an axe? Unlikely. Will it allow you to host services in a way that is considerably more robust and flexible than hosting them on a single box? Yes. -Steve -- If this were played upon a stage now, I could condemn it as an improbable fiction. - Fabian, Twelfth Night, III,v