Thread: Re: [GENERAL] Huge Pages - setting the right value
We are experiencing some performance issues because of high CPU load. So I would like to ask one more time. The exact question is: Does PostgreSQL can use huge pages for processes or only for shared buffers? (Does it make any sense to set the number of huge pages above the shared_buffers?) Any help or suggestions would be much appreciated! -- View this message in context: http://www.postgresql-archive.org/Huge-Pages-setting-the-right-value-tp5952972p5965956.html Sent from the PostgreSQL - general mailing list archive at Nabble.com.
I can't give you an absolutely authoritative answer, but because of the way hugepages are implemented and allocated, I can'tthink how they could be used for other processes. Linux hugepages are either 2m or 1g, far too large for any likelyprocesses to require. They cannot be allocated in partial pages. Sent from my iPad > On Jun 11, 2017, at 10:02 AM, pinker <pinker@onet.eu> wrote: > > We are experiencing some performance issues because of high CPU load. So I > would like to ask one more time. The exact question is: > Does PostgreSQL can use huge pages for processes or only for shared buffers? > (Does it make any sense to set the number of huge pages above the > shared_buffers?) > Any help or suggestions would be much appreciated! > > > > -- > View this message in context: http://www.postgresql-archive.org/Huge-Pages-setting-the-right-value-tp5952972p5965956.html > Sent from the PostgreSQL - general mailing list archive at Nabble.com. > > > -- > Sent via pgsql-general mailing list (pgsql-general@postgresql.org) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-general
Andrew Kerber wrote > I can't give you an absolutely authoritative answer, but because of the > way hugepages are implemented and allocated, I can't think how they could > be used for other processes. Linux hugepages are either 2m or 1g, far too > large for any likely processes to require. They cannot be allocated in > partial pages. thank you for your help. My system is using 2MB pages for shared buffers. I have checked and one of my processes has used 606788kB of memory, so potentially could use ~ 300 huge pages, but does postgres can use it for non shared memory? -- View this message in context: http://www.postgresql-archive.org/Huge-Pages-setting-the-right-value-tp5952972p5965963.html Sent from the PostgreSQL - general mailing list archive at Nabble.com.
I am sure it does not. Sent from my iPhone > On Jun 11, 2017, at 10:50 AM, pinker <pinker@onet.eu> wrote: > > Andrew Kerber wrote >> I can't give you an absolutely authoritative answer, but because of the >> way hugepages are implemented and allocated, I can't think how they could >> be used for other processes. Linux hugepages are either 2m or 1g, far too >> large for any likely processes to require. They cannot be allocated in >> partial pages. > > thank you for your help. > My system is using 2MB pages for shared buffers. I have checked and one of > my processes has used 606788kB of memory, so potentially could use ~ 300 > huge pages, but does postgres can use it for non shared memory? > > > > > -- > View this message in context: http://www.postgresql-archive.org/Huge-Pages-setting-the-right-value-tp5952972p5965963.html > Sent from the PostgreSQL - general mailing list archive at Nabble.com. > > > -- > Sent via pgsql-general mailing list (pgsql-general@postgresql.org) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-general
2017-06-12 7:52 GMT+12:00 Andrew Kerber <andrew.kerber@gmail.com>:
I am sure it does not.
Sent from my iPhone
> On Jun 11, 2017, at 10:50 AM, pinker <pinker@onet.eu> wrote:
>
> Andrew Kerber wrote
>> I can't give you an absolutely authoritative answer, but because of the
>> way hugepages are implemented and allocated, I can't think how they could
>> be used for other processes. Linux hugepages are either 2m or 1g, far too
>> large for any likely processes to require. They cannot be allocated in
>> partial pages.
>
> thank you for your help.
> My system is using 2MB pages for shared buffers. I have checked and one of
> my processes has used 606788kB of memory, so potentially could use ~ 300
> huge pages, but does postgres can use it for non shared memory?
>
>
>
>
> --
> View this message in context: http://www.postgresql-archive.org/Huge-Pages-setting-the- right-value-tp5952972p5965963. html
> Sent from the PostgreSQL - general mailing list archive at Nabble.com.
>
>
> --
> Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-general
--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
In my case, we had the HugePages enabled but not configured in our Master DB Server. When we increased the server resources (More RAM & CPU) we had lots of issues with HugePages. Specially I/O ones. Had to disabled it.
Running Ubuntu 14.04 Server @ Amazon.
Lucas
Was that transparent hugepages or standard hugepages? databases commonly have problems dealing with transparent hugepages.
On Sun, Jun 11, 2017 at 4:39 PM, Lucas Possamai <drum.lucas@gmail.com> wrote:
2017-06-12 7:52 GMT+12:00 Andrew Kerber <andrew.kerber@gmail.com>:I am sure it does not.
Sent from my iPhone
> On Jun 11, 2017, at 10:50 AM, pinker <pinker@onet.eu> wrote:
>
> Andrew Kerber wrote
>> I can't give you an absolutely authoritative answer, but because of the
>> way hugepages are implemented and allocated, I can't think how they could
>> be used for other processes. Linux hugepages are either 2m or 1g, far too
>> large for any likely processes to require. They cannot be allocated in
>> partial pages.
>
> thank you for your help.
> My system is using 2MB pages for shared buffers. I have checked and one of
> my processes has used 606788kB of memory, so potentially could use ~ 300
> huge pages, but does postgres can use it for non shared memory?
>
>
>
>
> --
> View this message in context: http://www.postgresql-archive.org/Huge-Pages-setting-the-rig ht-value-tp5952972p5965963.htm l
> Sent from the PostgreSQL - general mailing list archive at Nabble.com.
>
>
> --
> Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-general
--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general In my case, we had the HugePages enabled but not configured in our Master DB Server. When we increased the server resources (More RAM & CPU) we had lots of issues with HugePages. Specially I/O ones. Had to disabled it.Running Ubuntu 14.04 Server @ Amazon.Lucas
--
Andrew W. Kerber
'If at first you dont succeed, dont take up skydiving.'
'If at first you dont succeed, dont take up skydiving.'
2017-06-12 9:52 GMT+12:00 Andrew Kerber <andrew.kerber@gmail.com>:
Was that transparent hugepages or standard hugepages? databases commonly have problems dealing with transparent hugepages.
IN my case, it was the Transparent Hugepages....
Lucas
Yes, those should always be disabled using tuned or other methods. Using tuned is described here (second method). I think the grub.conf method described is unique to RHEL/OEL/CENTOS. http://houseofbrick.com/disabling-transparent-hugepages-using-tuned/
On Sun, Jun 11, 2017 at 5:00 PM, Lucas Possamai <drum.lucas@gmail.com> wrote:
2017-06-12 9:52 GMT+12:00 Andrew Kerber <andrew.kerber@gmail.com>:Was that transparent hugepages or standard hugepages? databases commonly have problems dealing with transparent hugepages.IN my case, it was the Transparent Hugepages....Lucas
--
Andrew W. Kerber
'If at first you dont succeed, dont take up skydiving.'
'If at first you dont succeed, dont take up skydiving.'
standard hugepages, transparent are disabled. They were set exactly following the procedure from postgres documentation. -- View this message in context: http://www.postgresql-archive.org/Huge-Pages-setting-the-right-value-tp5952972p5966064.html Sent from the PostgreSQL - general mailing list archive at Nabble.com.