[Gluster-devel] Wrong assumptions about disperse
jdarcy at redhat.com
Fri Jun 17 11:33:26 UTC 2016
> The math used by disperse, if tested alone outside gluster, is much
> faster than it seems. AFAIK the real problem of EC is the communications
> layer. It adds a lot of latency and having to communicate simultaneously
> and coordinate 6 or more bricks has a big impact.
Thanks for posting this, Xavier. "Erasure codes are slow because of the
math" is a common misunderstanding, unfortunately sustained by much of
the marketing around various EC implementations. As far as I can tell,
EC is mostly fast or slow according to how well the implementation
solves coordination problems, except for repair (in which case the
choice of algorithm can affect how much data needs to be moved).
More information about the Gluster-devel