Re: Re: patch submission: truncate trailing nulls from heap rows to reduce the size of the null bitmap [Review]

Sorry for the late response, I just happened to see this yesterday.

Running a general benchmark against the patch as Keven suggests is a good idea. 

Amit, can you supply the actual values you saw when running pgbench (the 3 values for each run)? I'd like to verify that the 1% difference isn't due to some file system/OS variability (would be interested in what the stdev is for the values). Also, do you happen to have some information about the hardware you ran on?

Meanwhile, I'll rerun on my end to see if I can reproduce your numbers. And I'll get a run of pgbench with minimal I/O since that can induce a lot of variability.

Also, Amit, I want to thank you for all your work on this patch, it is greatly appreciated.

Thanks

-Jamie

On Monday, December 24, 2012 7:58 PM Kevin Grittner wrote:
Simon Riggs wrote:
>Not really sure about the 100s of columns use case. But showing gain in useful places in these more common cases wins
my vote. Thanks for testing. Barring objections, will commit.
>Do we have any results on just a plain, old stock pgbench run, with
the default table definitions? That would be a reassuring complement to the other tests.
Sever Configuration:
The database cluster will be initialized with locales COLLATE: C CTYPE: C MESSAGES: en_US.UTF-8 MONETARY: en_US.UTF-8 NUMERIC: en_US.UTF-8 TIME: en_US.UTF-8 shared_buffers = 1GB
checkpoint_segments = 255
checkpoint_timeout = 15min pgbench:
transaction type: TPC-B (sort of)
scaling factor: 75
query mode: simple
number of clients: 8
number of threads: 8
duration: 600 s Performance: Average of 3 runs of pgbench in tps
9.3devel | with trailing null patch
----------+--------------------------
578.9872 | 573.4980 With Regards,
Amit Kapila.

On Wednesday, January 23, 2013 2:30 AM Jameison Martin wrote:

> Sorry for the late response, I just happened to see this yesterday.

> Running a general benchmark against the patch as Keven suggests is a good
idea. 

> Amit, can you supply the actual values you saw when running pgbench (the 3
values for each run)?

I haven't saved that, but I have plan to run it once again as Simon also
pointed it.

> I'd like to verify
> that the 1% difference isn't due to some file system/OS variability (would
be interested in what the stdev is for the
> values). Also, do you happen to have some information about the hardware
you ran on?

Platform details:
Operating System: Suse-Linux 10.2 x86_64
Hardware : 4 core (Intel(R) Xeon(R) CPU L5408 @ 2.13GHz)
RAM : 24GB

> Meanwhile, I'll rerun on my end to see if I can reproduce your numbers.
And I'll get a run of pgbench with minimal I/O > since that can induce a lot
of variability.

> Also, Amit, I want to thank you for all your work on this patch, it is
greatly appreciated.

As a Reviewer, I tried my best for this patch.

On Monday, December 24, 2012 7:58 PM Kevin Grittner wrote:
Simon Riggs wrote:

Not really sure about the 100s of columns use case.

But showing gain in useful places in these more common cases wins
my vote.

Thanks for testing. Barring objections, will commit.
Do we have any results on just a plain, old stock pgbench run, with
the default table definitions?

That would be a reassuring complement to the other tests.
Sever Configuration:
The database cluster will be initialized with locales
COLLATE: C
CTYPE: C
MESSAGES: en_US.UTF-8
MONETARY: en_US.UTF-8
NUMERIC: en_US.UTF-8
TIME: en_US.UTF-8

shared_buffers = 1GB
checkpoint_segments = 255
checkpoint_timeout = 15min

pgbench:
transaction type: TPC-B (sort of)
scaling factor: 75
query mode: simple
number of clients: 8
number of threads: 8
duration: 600 s

Performance: Average of 3 runs of pgbench in tps
9.3devel | with trailing null patch
----------+--------------------------
578.9872 | 573.4980

With Regards,
Amit Kapila.

On Wednesday, January 23, 2013 2:30 AM Jameison Martin wrote:

> Sorry for the late response, I just happened to see this yesterday.

> Running a general benchmark against the patch as Keven suggests is a good
idea. 

> Amit, can you supply the actual values you saw when running pgbench (the 3
values for each run)? I'd like to verify
> that the 1% difference isn't due to some file system/OS variability (would
be interested in what the stdev is for the
> values). Also, do you happen to have some information about the hardware
you ran on?

Performance data for 5 runs is as below:

System Configuration:
Hardware : 4 core (Intel(R) Xeon(R) CPU L5408 @ 2.13GHz) & RAM : 24GB
Operating System: Suse-Linux 10.2 x86_64

Sever Configuration:
The database cluster will be initialized with locales
COLLATE: C
CTYPE: C
MESSAGES: en_US.UTF-8
MONETARY: en_US.UTF-8
NUMERIC: en_US.UTF-8
TIME: en_US.UTF-8

shared_buffers = 2GB
checkpoint_segments = 255
checkpoint_timeout = 10min

pgbench:
transaction type: TPC-B (sort of)
scaling factor: 75
query mode: simple
number of clients: 8
number of threads: 8
duration: 300 s

original patch
Run-1: 579.596730 570.212601
Run-2: 577.220325 576.719402
Run-3: 571.792736 574.118542
Run-4: 573.376879 571.548136
Run-5: 573.469166 576.321368

Avg : 575.091167 573.784009

With Regards,
Amit Kapila.

On Monday, December 24, 2012 7:58 PM Kevin Grittner wrote:
Simon Riggs wrote:

Not really sure about the 100s of columns use case.

But showing gain in useful places in these more common cases wins
my vote.

Thanks for testing. Barring objections, will commit.
Do we have any results on just a plain, old stock pgbench run, with
the default table definitions?

That would be a reassuring complement to the other tests.
Sever Configuration:
The database cluster will be initialized with locales
COLLATE: C
CTYPE: C
MESSAGES: en_US.UTF-8
MONETARY: en_US.UTF-8
NUMERIC: en_US.UTF-8
TIME: en_US.UTF-8

shared_buffers = 1GB
checkpoint_segments = 255
checkpoint_timeout = 15min

pgbench:
transaction type: TPC-B (sort of)
scaling factor: 75
query mode: simple
number of clients: 8
number of threads: 8
duration: 600 s

Performance: Average of 3 runs of pgbench in tps
9.3devel | with trailing null patch
----------+--------------------------
578.9872 | 573.4980

With Regards,
Amit Kapila.