Solutions

The pages here describe one deployment for each kind of workload and explain every choice.

Each page below covers one workload and describes a cluster designed for it. The page lists which Storage Class each part of the data uses, how many Chunkservers the design needs, and what the choice costs. The figures are round numbers chosen for the example. They show the reasoning that connects them, which you can repeat with your own numbers.

Virtualization covers two workloads with opposite demands on one cluster: disk images of running guests, and the template and backup library that never changes. CDN and media covers the origin behind a content delivery network: a library written once and read many times, and what a second site needs. Backup target covers a landing zone for nightly backups, where the data is written once at night and read only when something has to be restored. Computation on storage nodes covers a cluster whose machines also run the jobs that read the data, and how to keep a read on the machine that holds the copy. Two data rooms covers a cluster divided into locations PRO (PRO Edition only), each of which keeps its own redundancy.

None of these pages calculates the size of a cluster from scratch. Planning & Sizing covers that work, which starts from the workload and ends with the hardware. The pages here cover decisions that have to be justified rather than calculated. Earlier pages answer whether MooseFS suits a workload at all: where it fits and where it does not.