Hi!
On Jul 14, 2008, at 6:32 AM, Jay Pipes wrote:
Clustered index organizations (especially ones with page-level block
organization) generate a hot-spot in memory and on-disk when
auto-incrementing primary keys are used. Essentially, with an
This is why we were discussing flipping the bits on UUID. AKA if we
flip the bits correctly we have a MAC for the beginning of the key.
So:
constant + timeseries
It is not quite this simple, and we would break the ISO for storage
(the end user would never see this), but we would keep increasing in
the same manner your see for AUTOINCREMENT.
Bigger picture problem is that users use UUID in order to solve the
distributed problem. There was a "short UUID" which was introduced in
5.1 I believe but it has the gotcha that it uses server id in order to
stay "unique". Problem with this? How many times have you seen a site
botch up their master id?
While writing this I am also wondering if the UUID we are discussing
will solve the "seen that" problem in circular replication.... since
we have a high water mark, and we can distinguish based on MAC
address...
Cheers,
-Brian
--
_______________________________________________________
Brian "Krow" Aker, brian at tangent.org
Seattle, Washington
http://krow.net/ <-- Me
http://tangent.org/ <-- Software
_______________________________________________________
You can't grep a dead tree.
_______________________________________________
Mailing list: https://launchpad.net/~drizzle-discuss
Post to : [email protected]
Unsubscribe : https://launchpad.net/~drizzle-discuss
More help : https://help.launchpad.net/ListHelp
