[tahoe-dev] Brewer's CAP theorem

David-Sarah Hopwood david-sarah at jacaranda.org
Thu Aug 20 04:19:04 UTC 2009


Brian Warner wrote:
> [...] (I wonder if there's a sort of Meyers-Brigg
> personality test that instead looks at your feelings about Brewer's CAP
> theorem: apparently I've been landing on the "prefers availability over
> consistency" side all this time. You could probably interpret the CAP
> theorem as a sort of consistency-vs-availability-vs-reliability thing,
> and apparently I've been on the availability-over-reliability side too.
> But I digress).

My feeling about the CAP theorem: it should instead be called the
"Serializability, Availability, Partition-Tolerance" theorem. That is,
it is using a very strong definition of consistency, which is more
accurately called serializability.

"Eventual consistency", on the other hand, which many people advocate
as an alternative to serializability, is not very well defined at all.
For example <http://queue.acm.org/detail.cfm?id=1466448> seems to use
it to refer to anything between causal consistency (rather strong)
and monotonic-write consistency (extremely weak), which is a very wide
range. I question whether it makes any sense at all to refer to this
range of properties by a single term.

-- 
David-Sarah Hopwood  ⚥  http://davidsarah.livejournal.com




More information about the tahoe-dev mailing list