Re: questions about PG update performance

On 27/10/15 11:37, Jim Nasby wrote:
> On 10/25/15 9:36 PM, Kisung Kim wrote:
>> I want to explain for our clients that PG's update performance is
>> comparable to Oracle's.
>
> There's really only 2 ways you can answer that. You can either
> handwave the question away ("Yes, update performance is comparable."),
> or you have to do actual benchmarking. Trying to answer this from a
> theoretical standpoint is completely useless because there's an absurd
> number of things that will affect this:
>
> Number of columns
> Data types
> Size of overall transaction
> Percent of transactions that roll back
> Size of table
> What % of table is updated every day
> Underlying hardware
> What OS the database is running on
> What filesystem the database is running on
>
> ... and that's just off the top of my head.
>
> Or to look at it another way, I guarantee you can create a scenario
> where Postgres beats the pants off Oracle, *or vice versa*. So you
> have to either go with an answer along the lines of "For most
> workloads the performance of both databases is similar." or you have
> to benchmark the actual application in question. Most performance
> issues you find will probably be correctable with a moderate amount of
> work.
>
> To me, the real tradeoff between Postgres and Oracle (or any other
> commercial database) is whether you'd rather spend money on expert
> employees or software contracts.

And of course, on how you alter the tuning parameters in
postgresql.conf, like temp_buffers and work_mem. The 'correct' values
will depend on your workload and amount of RAM etc.