Re: Major features for 9.1

Andrew Lardinois <lardinois@gmail.com> wrote:

> This must be the article in ACM-SIGMOD:
>
http://portal.acm.org/citation.cfm?id=1376616.1376690&coll=DL&dl=GUIDE&CFID=16503930&CFTOKEN=26926199
> Which has a nice abstract, but does anybody on the list have the
> complete article?

I did pay for and download the article, but it is a copyrighted
work, so I can't share it.  In an email exchange with Michael Cahill
he gave permission for me to share the URL of his personal copy of
the work with the list.  See my post in the archives here:

http://archives.postgresql.org/message-id/4A0019EE.EE98.0025.0@wicourts.gov

Please note you are not allowed to distribute copies you download
from that URL either; it is restricted to the personal use of
whoever downloads it.

On the other hand, there is no such restriction on Michael Cahill's
later doctoral thesis, which covers everything in the ACM SIGMOD
paper and more.

> There was also a paper published all the way back in 1997 by the
> Information Processing Society of Japan entitled:
> "A concurrency control algorithm using serialization graph testing
> with write deferring"

Do you have a URL?  I haven't read that one, but I remember seeing
references to attempts to do full graph testing which failed due to
the high cost of doing so.  The main point of the SSI technique is
that it keys on a "dangerous structure" always found in cycles which
can cause anomalies (and from what I've seen rarely present without
causing such anomalies).  Testing for this two to three transaction
dangerous structure is much less expensive.

-Kevin