OneFS - High Performance Parallel I/O - page 145

of size N 0 + M where N 0 = 2N + M. These double-length protection groups
will \wrap" once over the same set of nodes, as they are laid out. Clearly,
since each protection group contains exactly two disks' worth of redundancy
(M = 2), this mechanism will allow for the double-drive failure. When a node
fails, it causes two erasures in the protection group, but again M = 2 will
provide double-disk redundancy.
Most important for small clusters, this method of striping is highly e-
cient, with an on-disk eciency of 1Mb(N + M). For example, on a cluster
of five nodes with double-failure protection, were one to use N = 3, M = 2,
and b = 1, one would obtain a 3+2 protection group with an eciency of 1 2
5
or 60%. Using the same 5-node cluster but with each protection group laid
out over b = 2 stripes, we could obtain 122(3 + 2) or 80% eciency on disk,
retaining the double-drive failure protection and sacrificing only double-node
failure protection.
OneFS supports erasure code protection levels of N + 1, N + 2, N + 3,
and N + 4 (all b = 1), as well as N + 2 : 1 (double-drive and single-node
tolerant, b = 2) and N + 3 : 1 (triple-drive and single-node tolerant, b = 3),
as well as mirroring levels between 2 (twice mirrored) and 8 (eight-times
mirrored). Protection levels can be applied to individual files, directories, and
their contents; or the entire system.
OneFS enables an administrator to modify the protection policy in real
time, while clients are attached and are reading and writing data. Note that
increasing a cluster's protection level may increase the amount of space con-
sumed by the data on the cluster.
11.6 Dynamic Scale/Scale on Demand
11.6.1 Performance and Capacity
In contrast to traditional storage systems that must \scale up" when ad-
ditional performance or capacity is needed, OneFS enables an Isilon storage
system to \scale out," seamlessly increasing the existing le system or volume
into petabytes of capacity while increasing performance in tandem in a linear
fashion.
Adding capacity and performance capabilities to an Isilon cluster is sig-
nicantly easier than with other storage systems|requiring only three simple
steps for the storage administrator: adding another node into the rack, attach-
ing the node to the Infiniband network, and instructing the cluster to add the
additional node. The new node provides additional capacity and performance
since each node includes CPU, memory, cache, network, NVRAM, and I/O
control pathways.