Re: Re: Re: [GSOC 18] Performance Farm Project——Initialization Project

From: Mark Wong <mark(at)2ndQuadrant(dot)com>
To: Dave Page <dpage(at)pgadmin(dot)org>
Cc: Hongyuan Ma <cs_maleicacid(at)163(dot)com>, "pgsql-hackers(at)postgresql(dot)org" <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Re: Re: [GSOC 18] Performance Farm Project——Initialization Project
Date: 2018-03-14 23:14:38
Message-ID: 20180314231438.GA16918@2ndQuadrant.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi,

I have some additional comments to a couple areas...

On Wed, Mar 14, 2018 at 05:33:00PM -0400, Dave Page wrote:
> On Tue, Mar 13, 2018 at 11:31 PM, Hongyuan Ma <cs_maleicacid(at)163(dot)com> wrote:
> > At the same time I hope you can help me understand the functional
> > requirements of this project more clearly. Here are some of my thoughts
> > on PerfFarm:
> >
> > - I see this comment in the client folder (In line 15 of the "pgperffarm
> > \ client \ benchmarks \ pgbench.py" file):
> > '''
> > # TODO allow running custom scripts, not just the default
> > '''
> > Will PerfFarm have many test items so that the test item search function
> > or even the test item sort function is expected to be provided?
> >
>
> I don't know - Tomas or Mark are probably the best folks to ask about that.
> I spent some time on the initial web app work, but then ran out of spare
> cycles.

Yes, there is potential for the number of tests to change over time.
(Maybe a test is less relevant, maybe there is a more useful test down
he road.)

I'm haven't thought too much about sorting and searching functions. But
right now there is just pgbench with 6 basic configurations: read/write,
read-only, and a low, medium, and high data set size for each of those.

> > - What value will be used to determine if a machine's performance is
> > improving or declining?
> >
>
> Good question - I think that needs discussion, as it's not necessarily
> clear-cut when you think that performance may vary at different numbers of
> clients.

My thought here is that each test should provide a single metric,
defined by the test itself, so expect this to be define in the test
results sent to the web site.

I do expect that off of the test results should be archived somehow and
retrievable. But the idea I have is to have a generic high level view
that all of the different tests can share.

> > - I see BuildFarm assigning an animal name to each registered machine. Will
> > PerfFarm also have this interesting feature?
> >
>
> It was going to, but I forget what we decided on for a naming scheme!
> Another discussion item I think - in the code, we can just use "Name" or
> similar.

Since the buildfarm is composed of animals, I thought plants would be a
complimentary scheme? I'm also not sure if this was discussed
previously...

Regards,
Mark

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tom Lane 2018-03-14 23:31:08 Instability in parallel regression tests
Previous Message PG Bug reporting form 2018-03-14 23:12:26 BUG #15112: Unable to run pg_upgrade with earthdistance extension