Simon Riggs wrote:
> On Tue, 2007-03-27 at 20:48 -0400, Tom Lane wrote:
>> "Simon Riggs" <simon@2ndquadrant.com> writes:
>>> It seems possible to reduce overall WAL volume by roughly 25% on common
>>> workloads by optimising the way UPDATE statements generate WAL.
>> This seems a huge amount of work to optimize *one* benchmark.
>
> Please don't beat me with that. I wouldn't suggest it if I didn't think
> it would help real users. The analysis of the WAL volume was done using
> a benchmark, but only as a guide to indicate likely usage patterns.
> There aren't many real world heavy UPDATE scenarios to analyze right now
> because people have previously actively avoided such usage.
>
>> If it
>> weren't so narrowly focused on the properties of a particular benchmark
>> (mostly UPDATE, mostly a few columns in wide tuples), I'd get more
>> excited.
>
As a reference unless there is some further restriction I'm not
understanding I've seen a lot of scenarios with this profile.
An online multiplayer game, around 60,000 active users (out of 250k
registered) did 6m page view per day, most dynamic.
Most of the interactions between players, or between players and objects
in the system resulted in updates to tables perhaps 512-1kb wide. A fair
number of strings etc.
Session information was tracked for active users, similar description
but even more updates.
I'd echo the other poster, that at the time we didn't fully know
postgresql's performance profile (this was 6 years ago now) to code to,
and ended up using MySQL because it worked and deadlines were tight.
I can think of a lot of other scenarios as well that would be helped.
Echoing another poster, it would be nice to round of a couple of the
gotcha edges. This may be one of them.
- August
