Jay Pipes wrote:
Alex Yurchenko wrote:
On Thu, 20 Aug 2009 12:36:42 -0400, Morgan Tocker <[email protected]>
wrote:
While I agree with the comments about CRCs being weak, they are
awesomely fast. And Nehalem CPUs make them even faster:
http://en.wikipedia.org/wiki/SSE4#SSE4.2
Well, according to http://www.cryptopp.com/benchmarks.html CRC32
might be
awesomely fast only on those Nehalem things. Otherwise it seems to be
pathetically slow (for a 32bit value).
General unsuitability for the task was already addressed in previous
posts.
Baron has chosen FNV hash not for naught.
The next argument is just which sane default should be chosen. In
which case, I'd be interested to let benchmarks speak.
All who responded to this thread:
Thank you! I'll be putting together a wiki page which summarizes the
viewpoints expressed and lays out (hopefully cleanly) a set of options
to either build into the server, the replication streaming, both, or
none :)
Hmmm. How much consideration has been given to cross engine issues? A
checksum based on simple byte stream is going to give different results
on a record in an engine that stores in primary key order and an engine
that stores in physical order.
--
Jim Starkey
Sent from Shearwater, off
the coast of New England
_______________________________________________
Mailing list: https://launchpad.net/~drizzle-discuss
Post to : [email protected]
Unsubscribe : https://launchpad.net/~drizzle-discuss
More help : https://help.launchpad.net/ListHelp
