building a server

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
5 messages Options
lup
Reply | Threaded
Open this post in threaded view
|

building a server

lup
So I revisited https://momjian.us/main/writings/pgsql/hw_selection.pdf
Near the end (page 24) I spotted: 
Heavy use of server-side functions might generate significant CPU load

Just want to confirm that this referring to workload on the mainly-single-threaded server not an advocacy for more cores.

Thanks

PS The wiki page at https://wiki.postgresql.org/wiki/Database_Hardware lists a reference to a PDF by Greg Smith.  The url vectors off into some proprietary e-book enticement. Is that intentional?

Reply | Threaded
Open this post in threaded view
|

Re: building a server

John R Pierce
On 12/3/2017 12:47 PM, Gmail wrote:
So I revisited https://momjian.us/main/writings/pgsql/hw_selection.pdf
Near the end (page 24) I spotted: 
Heavy use of server-side functions might generate significant CPU load

Just want to confirm that this referring to workload on the mainly-single-threaded server not an advocacy for more cores.

well, if you have significant concurrency with high CPU usage, you want both more AND faster cores


-- 
john r pierce, recycling bits in santa cruz
lup
Reply | Threaded
Open this post in threaded view
|

Re: building a server

lup



On 12/03/2017 02:15 PM, John R Pierce wrote:
On 12/3/2017 12:47 PM, Gmail wrote:
So I revisited https://momjian.us/main/writings/pgsql/hw_selection.pdf
Near the end (page 24) I spotted: 
Heavy use of server-side functions might generate significant CPU load

Just want to confirm that this referring to workload on the mainly-single-threaded server not an advocacy for more cores.

well, if you have significant concurrency with high CPU usage, you want both more AND faster cores


-- 
john r pierce, recycling bits in santa cruz
Granted!  I suppose I'm looking for confirmation/correction on believing that for the server to make use of multiple cores is more dependent on the nature of the queries handled.  Concurrency should not be a huge problem for this project, though I've put pieces in place to mitigate that issue.
Reply | Threaded
Open this post in threaded view
|

Re: building a server

John R Pierce
On 12/3/2017 1:39 PM, Rob Sargent wrote:
> Granted! I suppose I'm looking for confirmation/correction on
> believing that for the server to make use of multiple cores is more
> dependent on the nature of the queries handled.  Concurrency should
> not be a huge problem for this project, though I've put pieces in
> place to mitigate that issue.


a single connection will only ever use more than one core if its
explicitly doing concurrent query.


--
john r pierce, recycling bits in santa cruz


Reply | Threaded
Open this post in threaded view
|

Re: building a server

hvjunk
In reply to this post by lup

On 03 Des. 2017, at 23:39 , Rob Sargent <[hidden email]> wrote:
On 12/03/2017 02:15 PM, John R Pierce wrote:
On 12/3/2017 12:47 PM, Gmail wrote:
So I revisited https://momjian.us/main/writings/pgsql/hw_selection.pdf
Near the end (page 24) I spotted: 
Heavy use of server-side functions might generate significant CPU load

Just want to confirm that this referring to workload on the mainly-single-threaded server not an advocacy for more cores.

well, if you have significant concurrency with high CPU usage, you want both more AND faster cores

john r pierce, recycling bits in santa cruz

Granted!  I suppose I'm looking for confirmation/correction on believing that for the server to make use of multiple cores is more dependent on the nature of the queries handled.  Concurrency should not be a huge problem for this project, though I've put pieces in place to mitigate that issue.

I would not say that much on the “nature of the queries”, but more on the “number of concurrent sessions doing queries”. Granted that with the later postgresql versions, given you’ve set the needed to enable parallel queries and you have “big-enough-tables” some selects does make use of multiple cores.

As they always say: YMMV