On Wed, Nov 23, 2016 at 11:35:38PM -0800, Peter Geoghegan wrote:
> On Wed, Nov 23, 2016 at 11:32 PM, Tsunakawa, Takayuki
> <tsunakawa.takay@jp.fujitsu.com> wrote:
> > IMHO, overall, there should be pros and cons of the current approach and the new UNDo one (like Oracle?), depending
onthe workload. Under update-heavy workload, the UNDO method may be better. OTOH, under the mostly-INSERT workload
(likedata warehouse?), the current method will be better because it writes no log for UNDO.
>
> I believe that you are correct about that.
This is probably similar to our use of double-buffering, using
shared_buffers and the kernel cache --- in some cases, the
double-buffering is better (variable workloads), while in other cases a
single cache is best.
We have had trouble figuring out if we need to support both single and
double caching methods, and when to recommend one over the other. Seems
UNDO would have the same complexity.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +