Re: Major features for 9.1

Joshua Berkus wrote:

> SSI: PostgreSQL first

First *production* implementation. The 2008 ACM SIGMOD paper
included a prototype implementation in Oracle Berkeley DB, and
Michael Cahill's 2009 doctoral thesis included a prototype
implementation in InnoDB. Even though Cahill was hired by Oracle in
2009, I haven't been able to find any indication that Oracle has
included this technology in any production release of either of these
products, or any other. I've checked their documentations without
finding references to anything but S2PL based SERIALIZABLE
implementations.

Having also tediously gone through every result page in in several
google searches using likely words, and diving into all pages which
look like they *could* refer to SSI, I haven't found any that
actually *do*.

I just wanted to provide enough detail to fashion a strong statement
that someone couldn't pick apart on the basis of the prototypes from
the academic papers.

-Kevin

On Sun, 10 Apr 2011 11:12:24 -0500, "Kevin Grittner"
<Kevin(dot)Grittner(at)wicourts(dot)gov> wrote:
> Joshua Berkus wrote:
>
>> SSI: PostgreSQL first
>
> First *production* implementation. The 2008 ACM SIGMOD paper
> included a prototype implementation in Oracle Berkeley DB, and
> Michael Cahill's 2009 doctoral thesis included a prototype
> implementation in InnoDB. Even though Cahill was hired by Oracle in
> 2009, I haven't been able to find any indication that Oracle has
> included this technology in any production release of either of these
> products, or any other. I've checked their documentations without
> finding references to anything but S2PL based SERIALIZABLE
> implementations.
>
> Having also tediously gone through every result page in in several
> google searches using likely words, and diving into all pages which
> look like they *could* refer to SSI, I haven't found any that
> actually *do*.
>
> I just wanted to provide enough detail to fashion a strong statement
> that someone couldn't pick apart on the basis of the prototypes from
> the academic papers.

I think you made our own argument though. We are releasing a "production"
version of SSI. Nobody else has done that.

JD

>
> -Kevin

--
PostgreSQL - XMPP: jdrake(at)jabber(dot)postgresql(dot)org
Consulting, Development, Support, Training
503-667-4564 - http://www.commandprompt.com/
The PostgreSQL Company, serving since 1997

On 04/10/2011 12:12 PM, Kevin Grittner wrote:
> Having also tediously gone through every result page in in several
> google searches using likely words, and diving into all pages which
> look like they *could* refer to SSI, I haven't found any that
> actually *do*.
>

I just fixed http://en.wikipedia.org/wiki/SSI to include this acronym,
and the page I found to point it toward:
http://en.wikipedia.org/wiki/Snapshot_isolation#Making_Snapshot_Isolation_Serializable
doesn't suggest it's anything other than a theoretical concept so far.
I think there's a very real academic possibility for your work here
too. I'd be writing to the ACM about your upcoming paper "Implementing
Serializable Snapshot Isolation in a MVCC Database" if I were you.
After 9.1 ships, of course. (Be good to update the Wikipedia pages on
this at that point, too, once there's a long-term documentation URL for 9.1)

--
Greg Smith 2ndQuadrant US greg(at)2ndQuadrant(dot)com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us

Greg Smith <greg(at)2ndQuadrant(dot)com> wrote:

> I just fixed http://en.wikipedia.org/wiki/SSI to include this
> acronym, and the page I found to point it toward:
>
http://en.wikipedia.org/wiki/Snapshot_isolation#Making_Snapshot_Isolation_Serializable
> doesn't suggest it's anything other than a theoretical concept so
> far.

Thanks for taking care of the SSI page. I've had my eye on that,
the page you pointed it to, and this one:

http://en.wikipedia.org/wiki/Isolation_%28database_systems%29

That last is pretty SS2PL-oriented, although it has a token
reference to snapshot isolation. Those two pages both probably need
significant work to eliminate that bias and properly reference the
fact that SSI (or as Wikipedia refers to it, SerializableSI) has
moved from an academic concept to a production product.

I've been a little concerned about updating those pages because I
know they have been whacked around to support different biases over
the last few years, and I didn't want to get into some battle over
it; so on the 12th of April I tried to initiate a discussion on the
topic on the Discussion tab of the page you linked to. So far there
has been no response, so perhaps those who were so eager to battle
it out on those pages have moved on and won't raise a fuss. At a
minimum they should be able to see that I made the effort to discuss
it before making the changes.

> I think there's a very real academic possibility for your work
> here too. I'd be writing to the ACM about your upcoming paper
> "Implementing Serializable Snapshot Isolation in a MVCC Database"
> if I were you.

I'm leaving that to Dan. I'm not in a "publish or perish"
environment, nor at this point even looking to build cred for
consulting gigs. Academic papers are not really my forte anyway.

> (Be good to update the Wikipedia pages on this at that point, too,
> once there's a long-term documentation URL for 9.1)

Yeah, the link to the PostgreSQL docs on the page you pointed to
still references the 8.2 docs. I wonder whether it's better form to
link to the 9.1 docs once they're out, or to link to the "current"
URL, so that it's always automatically pointing to the latest
version.

-Kevin

On 04/30/2011 11:52 AM, Kevin Grittner wrote:
> Yeah, the link to the PostgreSQL docs on the page you pointed to
> still references the 8.2 docs. I wonder whether it's better form to
> link to the 9.1 docs once they're out, or to link to the "current"
> URL, so that it's always automatically pointing to the latest
> version.
>

The 9.1 ones once they are out. Having a URL whose content changes
significantly over time drives some of the Wikipedia crowd crazy.

I wouldn't even get into arguing over the main page. I'd just add
"PostgreSQL 9.1 includes a true serialization implementation based on
the SerializableSI approach", add a footnote linking to the PostgreSQL
docs and your design page on the wiki, and call it a day. Let the
revert geeks fight over how the tone of the page needs to be adjusted to
reflect this new referenced fact.

--
Greg Smith 2ndQuadrant US greg(at)2ndQuadrant(dot)com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us

Greg Smith <greg(at)2ndQuadrant(dot)com> wrote:

> Having a URL whose content changes significantly over time drives
> some of the Wikipedia crowd crazy.

While I can understand the reason, there is a certain irony in that.

Thanks for the Wikipedia advice.

-Kevin