Sizing the cluster

What to work out before you buy hardware, in the order in which the answers depend on each other.

Start from the workload covers the terms in which to describe an installation, because the machines depend on what applications do to the data rather than on how much data there is. Choosing between copies and Erasure Coding covers the decision that determines what redundancy costs in space, in servers, and in the time a read takes. Chunkservers needed for an Erasure Coding scheme covers the two server counts that a scheme has to meet.

Two pages calculate figures from those decisions. Master Server RAM from the object count sizes the machine that holds the metadata. The RAM of that machine is counted from objects and chunks rather than from any amount of storage. Raw space and user space calculates the space that files can occupy from the raw space that you buy.

Hardware recommendations cover what to buy for each role, once these figures are known.