Writeback Cache

Alex Aizman, CTO of Nexenta, will be giving a talk on ZFS Writeback Cache at the OpenZFS Developer Summit 2015:

Writeback caching (aka write-behind caching) is the capability to write data to fast persistent cache, with subsequent - delayed, deferred and totally transparent for user - migration of this written data to backing store. In addition, the cache is required to perform as a tier 0 (zero) or fast tier - the latter distinguishes writeback caches from conventional (and ZFS own) write logs.

At Nexenta we designed, implemented, stabilized, debugged, stress-and-functionally tested, and optimized this new ZFS capability: writeback caching. We have introduced a new within-pool class of storage, a so called special vdev and the corresponding metaslab class, to support what’s the industry calls automated within-LUN (sub-LUN) tiering.

• Do users need a hybrid and truly tiered ZFS pool and why not to simply use all-flash?
• How does automated migration work versus any/all concurrent scenarios, user workloads, snapshot and dataset lifecycle events, hardware configurations?
• What are the apples-to-apples performance comparisons with ZFS own slog and L2ARC?

These and other questions will be answered.