Gilberto Castillo Martínez<gilberto(dot)castillo(at)etecsa(dot)cu> wrote:
>> > when it comes to SSI I am not finding any explanations
>> > for this acronym.
>>
>> http://wiki.postgresql.org/wiki/Serializable
>>
>> It's about a technique for serializable transactions which
>> doesn't add any blocking beyond what is present in snapshot
>> isolation. In contrast, most serializable implementations which
>> provide true serializability do this by blocking conflicting
>> transactions. A description of the SSI technique was first
>> published in the 2008 ACM SIGMOD, with independent confirmation
>> of some of the published performance improvements compared to
>> traditional S2PL serializability by an ACM committee. While two
>> prototype implementations were developed to support academic
>> papers on the topic, we believe this is the first production-
>> quality implementation. It includes many innovations which were
>> not described in any academic papers on the topic.
>
> Excelent!
>
> The News:
> It's about a technique for serializable transactions which doesn't
> add any blocking beyond what is present in snapshot isolation. It
> includes many innovations which were not described in any academic
> papers on the topic.
I didn't intend for that to be my recommended "elevator talk"
summary; I was just trying to quickly explain the feature to someone
who knew nothing of it. [pause] Oh. [pause] I guess maybe this
should be my starting point, although I usually like to harp on the
fact that it allows programmers to enforce business rules within
triggers or even application code without worrying about explicit
locking and how this can boost programming productivity, especially
in large shops.
I'm not sure I can hit both angles in one two-sentence summary.
Which angle is better? Maybe I could creep to three sentences to
allow both points?
-Kevin