ORACLE ON AIX PERFORMANCE PROBLEMS

[agis]

Hi to all,

I have an RS/6000 with2 CPUs and 2048 MB of RAM running tow instances
of ORACLE 8.1.7 database.One with 440MB SGA and with about 20
dedicated connections and about 100 in shared servers and one with
35MB SGA with almost no activity on JFS file system.

Some times on peak hours we face performance problems in response
time. Oracle is tuned and with vmstat i realized that we had a lot of
paging and swapping. I use nmon utility with the folwoing output :

% Used 100.0% 31.1% to Paging Space 12.0 29.0
numperm 73.1%
% Free 0.0% 68.9% to File System 457.6 815.3
minperm 20.0%
MB Used 2047.5MB 636.2MB Page Scans 407.7
maxperm 79.8%
MB Free 0.5MB 1411.8MB Page Cycles 0.0
minfree 120
Total(MB) 2048.0MB 2048.0MB Page Reclaim 0.0
maxfree 128

Also i post 2 vmstat outputs one with normal workload and one with the
performance problems.

NORMAL


kthr memory page faults cpu
----- ----------- ------------------------ ------------ -----------
r b avm fre re pi po fr sr cy in sy cs us sy id wa
2 2 178921 121 0 4 3 317 290 0 194 318 222 22 9 60 9
2 2 178921 128 0 12 5 113 159 0 819 6716 2565 63 16 13 7
2 2 178921 126 0 3 3 89 122 0 797 5971 2414 60 14 19 7
1 2 178921 123 0 2 1 50 75 0 477 2566 993 36 4 53 7
1 2 178921 125 0 2 3 73 120 0 549 2662 1243 45 5 40 10

PROBLEM

kthr memory page faults cpu
----- ----------- ------------------------ ------------ -----------
r b avm fre re pi po fr sr cy in sy cs us sy id wa
2 2 178979 123 0 4 3 317 290 0 194 322 224 22 9 60 9
2 2 178979 124 0 12 27 792 1973 0 1071 4877 3069 63 15 4 18
4 3 179532 112 0 66 37 889 1561 0 1150 3573 4763 72 17 2 9
3 3 179532 119 0 13 13 728 1319 0 1072 3792 4546 73 16 2 9
4 3 179532 124 0 9 65 733 1487 0 1135 4299 4041 66 21 3 10
5 2 179812 92 0 14 72 891 1666 0 1175 4183 3866 73 15 2 10
5 3 179881 123 0 10 156 851 3283 0 1153 2445 2031 36 8 10 45
2 4 179896 0 0 15 182 947 2160 0 1250 2719 2428 31 10 7 52
1 5 180565 32 0 7 189 596 2143 0 966 1563 1478 14 6 8 72
3 4 180655 107 0 53 104 688 1749 0 1148 4207 3253 60 13 2 25
3 4 179879 906 0 17 129 625 1167 0 1078 2961 3994 56 13 3 28
1 5 180201 0 0 50 107 765 1541 0 1102 2609 2824 29 9 8 54
1 4 181176 94 0 19 167 961 2153 0 1106 2439 2070 26 9 5 60
6 5 181338 117 0 87 59 633 1408 0 1138 6255 3419 63 14 2 21
2 4 181338 0 0 26 129 639 1952 0 1009 2332 1875 38 7 11 44
3 2 180957 125 0 50 130 403 940 0 915 2771 1677 33 7 9 51

I have determinate that we use about 1.4 GB for file caching. I want
to try reduce minperm and maxperm in order to allow at max about 30%
of RAM for file caching and i think i should gain 700MB free memory
for oracle.

Is my thought correct ?

Has someone faced the same problem ?

Is anything else i can do ?



Thanks in abvance and sorry for the extened thread.

Agis

ORACLE DBA

[Jurjen Oskam]

In article <d2aa3279.0309150615.4c673a1@posting.google.com> , agis wrote:

> NORMAL
>
> kthr memory page faults cpu
> ----- ----------- ------------------------ ------------ -----------
> r b avm fre re pi po fr sr cy in sy cs us sy id wa
> 2 2 178921 121 0 4 3 317 290 0 194 318 222 22 9 60 9
> 2 2 178921 128 0 12 5 113 159 0 819 6716 2565 63 16 13 7
> 2 2 178921 126 0 3 3 89 122 0 797 5971 2414 60 14 19 7
> 1 2 178921 123 0 2 1 50 75 0 477 2566 993 36 4 53 7
> 1 2 178921 125 0 2 3 73 120 0 549 2662 1243 45 5 40 10

Although you don't experience any user complaints here, this is also
problematic. Your free list is constantly at the low watermark, which
causes the page stealer to kick in and free a few frames. You also have
paging activity to/from paging space, which you don't want.

In the problem case you mention below, the same is basically happening,
only more so.

Some suggestions:

- reduce minperm / maxperm to about 10 / 15 % respectively
- set maxfree to 256

This way, you'll free up filesystem cache in return for computational
memory, and if the page stealer *does* kick in, it'll free a relatively
large amount of pages in one go. This is more efficient that starting,
freeing 8 pages, and immediately restarting because the 8 pages that
were just freed are already gone.

> Is anything else i can do ?

In Oracle, do you have the LOCK_SGA parameter defined? This causes the
SGA to be pinned in memory so that it can never be swapped out. If the
SGA is sized correctly, this will ensure that at least the SGA
will remain in "real" memory.

--
Jurjen Oskam
(Posted using Google since my ISP newsserver doesn't work right now)

PGP Key available at [url]http://www.stupendous.org/[/url]

[Geert]

Try: /usr/samples/kernel/vmtune -p10 -P30
It's a good start to tune your memory usage.


regards,
Geert


"agis" <agstamy@yahoo.com> schreef in bericht
news:d2aa3279.0309150615.4c673a1@posting.google.co m...

> Hi to all,
>
> I have an RS/6000 with2 CPUs and 2048 MB of RAM running tow instances
> of ORACLE 8.1.7 database.One with 440MB SGA and with about 20
> dedicated connections and about 100 in shared servers and one with
> 35MB SGA with almost no activity on JFS file system.
>
> Some times on peak hours we face performance problems in response
> time. Oracle is tuned and with vmstat i realized that we had a lot of
> paging and swapping. I use nmon utility with the folwoing output :
>
> % Used 100.0% 31.1% to Paging Space 12.0 29.0
> numperm 73.1%
> % Free 0.0% 68.9% to File System 457.6 815.3
> minperm 20.0%
> MB Used 2047.5MB 636.2MB Page Scans 407.7
> maxperm 79.8%
> MB Free 0.5MB 1411.8MB Page Cycles 0.0
> minfree 120
> Total(MB) 2048.0MB 2048.0MB Page Reclaim 0.0
> maxfree 128
>
> Also i post 2 vmstat outputs one with normal workload and one with the
> performance problems.
>
> NORMAL
>
>
> kthr memory page faults cpu
> ----- ----------- ------------------------ ------------ -----------
> r b avm fre re pi po fr sr cy in sy cs us sy id wa
> 2 2 178921 121 0 4 3 317 290 0 194 318 222 22 9 60 9
> 2 2 178921 128 0 12 5 113 159 0 819 6716 2565 63 16 13 7
> 2 2 178921 126 0 3 3 89 122 0 797 5971 2414 60 14 19 7
> 1 2 178921 123 0 2 1 50 75 0 477 2566 993 36 4 53 7
> 1 2 178921 125 0 2 3 73 120 0 549 2662 1243 45 5 40 10
>
> PROBLEM
>
> kthr memory page faults cpu
> ----- ----------- ------------------------ ------------ -----------
> r b avm fre re pi po fr sr cy in sy cs us sy id wa
> 2 2 178979 123 0 4 3 317 290 0 194 322 224 22 9 60 9
> 2 2 178979 124 0 12 27 792 1973 0 1071 4877 3069 63 15 4 18
> 4 3 179532 112 0 66 37 889 1561 0 1150 3573 4763 72 17 2 9
> 3 3 179532 119 0 13 13 728 1319 0 1072 3792 4546 73 16 2 9
> 4 3 179532 124 0 9 65 733 1487 0 1135 4299 4041 66 21 3 10
> 5 2 179812 92 0 14 72 891 1666 0 1175 4183 3866 73 15 2 10
> 5 3 179881 123 0 10 156 851 3283 0 1153 2445 2031 36 8 10 45
> 2 4 179896 0 0 15 182 947 2160 0 1250 2719 2428 31 10 7 52
> 1 5 180565 32 0 7 189 596 2143 0 966 1563 1478 14 6 8 72
> 3 4 180655 107 0 53 104 688 1749 0 1148 4207 3253 60 13 2 25
> 3 4 179879 906 0 17 129 625 1167 0 1078 2961 3994 56 13 3 28
> 1 5 180201 0 0 50 107 765 1541 0 1102 2609 2824 29 9 8 54
> 1 4 181176 94 0 19 167 961 2153 0 1106 2439 2070 26 9 5 60
> 6 5 181338 117 0 87 59 633 1408 0 1138 6255 3419 63 14 2 21
> 2 4 181338 0 0 26 129 639 1952 0 1009 2332 1875 38 7 11 44
> 3 2 180957 125 0 50 130 403 940 0 915 2771 1677 33 7 9 51
>
> I have determinate that we use about 1.4 GB for file caching. I want
> to try reduce minperm and maxperm in order to allow at max about 30%
> of RAM for file caching and i think i should gain 700MB free memory
> for oracle.
>
> Is my thought correct ?
>
> Has someone faced the same problem ?
>
> Is anything else i can do ?
>
>
>
> Thanks in abvance and sorry for the extened thread.
>
> Agis
>
> ORACLE DBA

[Ralph]

[email]agstamy@yahoo.com[/email] (agis) wrote in message news:<d2aa3279.0309150615.4c673a1@posting.google.c om>...

> Hi to all,
>
> I have an RS/6000 with2 CPUs and 2048 MB of RAM running tow instances
> of ORACLE 8.1.7 database.One with 440MB SGA and with about 20
> dedicated connections and about 100 in shared servers and one with
> 35MB SGA with almost no activity on JFS file system.
>
> Some times on peak hours we face performance problems in response
> time. Oracle is tuned and with vmstat i realized that we had a lot of
> paging and swapping. I use nmon utility with the folwoing output :
>
> % Used 100.0% 31.1% to Paging Space 12.0 29.0
> numperm 73.1%
> % Free 0.0% 68.9% to File System 457.6 815.3
> minperm 20.0%
> MB Used 2047.5MB 636.2MB Page Scans 407.7
> maxperm 79.8%
> MB Free 0.5MB 1411.8MB Page Cycles 0.0
> minfree 120
> Total(MB) 2048.0MB 2048.0MB Page Reclaim 0.0
> maxfree 128
>
> Also i post 2 vmstat outputs one with normal workload and one with the
> performance problems.
>
> NORMAL
>
>
> kthr memory page faults cpu
> ----- ----------- ------------------------ ------------ -----------
> r b avm fre re pi po fr sr cy in sy cs us sy id wa
> 2 2 178921 121 0 4 3 317 290 0 194 318 222 22 9 60 9
> 2 2 178921 128 0 12 5 113 159 0 819 6716 2565 63 16 13 7
> 2 2 178921 126 0 3 3 89 122 0 797 5971 2414 60 14 19 7
> 1 2 178921 123 0 2 1 50 75 0 477 2566 993 36 4 53 7
> 1 2 178921 125 0 2 3 73 120 0 549 2662 1243 45 5 40 10
>
> PROBLEM
>
> kthr memory page faults cpu
> ----- ----------- ------------------------ ------------ -----------
> r b avm fre re pi po fr sr cy in sy cs us sy id wa
> 2 2 178979 123 0 4 3 317 290 0 194 322 224 22 9 60 9
> 2 2 178979 124 0 12 27 792 1973 0 1071 4877 3069 63 15 4 18
> 4 3 179532 112 0 66 37 889 1561 0 1150 3573 4763 72 17 2 9
> 3 3 179532 119 0 13 13 728 1319 0 1072 3792 4546 73 16 2 9
> 4 3 179532 124 0 9 65 733 1487 0 1135 4299 4041 66 21 3 10
> 5 2 179812 92 0 14 72 891 1666 0 1175 4183 3866 73 15 2 10
> 5 3 179881 123 0 10 156 851 3283 0 1153 2445 2031 36 8 10 45
> 2 4 179896 0 0 15 182 947 2160 0 1250 2719 2428 31 10 7 52
> 1 5 180565 32 0 7 189 596 2143 0 966 1563 1478 14 6 8 72
> 3 4 180655 107 0 53 104 688 1749 0 1148 4207 3253 60 13 2 25
> 3 4 179879 906 0 17 129 625 1167 0 1078 2961 3994 56 13 3 28
> 1 5 180201 0 0 50 107 765 1541 0 1102 2609 2824 29 9 8 54
> 1 4 181176 94 0 19 167 961 2153 0 1106 2439 2070 26 9 5 60
> 6 5 181338 117 0 87 59 633 1408 0 1138 6255 3419 63 14 2 21
> 2 4 181338 0 0 26 129 639 1952 0 1009 2332 1875 38 7 11 44
> 3 2 180957 125 0 50 130 403 940 0 915 2771 1677 33 7 9 51
>
> I have determinate that we use about 1.4 GB for file caching. I want
> to try reduce minperm and maxperm in order to allow at max about 30%
> of RAM for file caching and i think i should gain 700MB free memory
> for oracle.
>
> Is my thought correct ?
>
> Has someone faced the same problem ?
>
> Is anything else i can do ?
>
>
>
> Thanks in abvance and sorry for the extened thread.
>
> Agis
>
> ORACLE DBA

I would start with minperm = 10 & maxperm = 30, that will definitely
help your performance.

What version of AIX?

Do you have Asynch I/O enabled? Are you mounting the filesystem with
the Oracle tablespaces with Direct I/O ?

Ralph

[Holger van Koll]

> In Oracle, do you have the LOCK_SGA parameter defined? This causes the

> SGA to be pinned in memory so that it can never be swapped out.

That only works if oracle runs as root, doesnt it?
At least I see no way for a non-root task to pin memory.

Btw. its obsoleted

[url]http://www.orafaq.com/parms/parm369.htm[/url]

[agis]

[email]ralph.rye@norgetek.com[/email] (Ralph) wrote in message news:<57422359.0309151205.51e11360@posting.google. com>...

> [email]agstamy@yahoo.com[/email] (agis) wrote in message news:<d2aa3279.0309150615.4c673a1@posting.google.c om>...

> > Hi to all,
> >
> > I have an RS/6000 with2 CPUs and 2048 MB of RAM running tow instances
> > of ORACLE 8.1.7 database.One with 440MB SGA and with about 20
> > dedicated connections and about 100 in shared servers and one with
> > 35MB SGA with almost no activity on JFS file system.
> >
> > Some times on peak hours we face performance problems in response
> > time. Oracle is tuned and with vmstat i realized that we had a lot of
> > paging and swapping. I use nmon utility with the folwoing output :
> >
> > % Used 100.0% 31.1% to Paging Space 12.0 29.0
> > numperm 73.1%
> > % Free 0.0% 68.9% to File System 457.6 815.3
> > minperm 20.0%
> > MB Used 2047.5MB 636.2MB Page Scans 407.7
> > maxperm 79.8%
> > MB Free 0.5MB 1411.8MB Page Cycles 0.0
> > minfree 120
> > Total(MB) 2048.0MB 2048.0MB Page Reclaim 0.0
> > maxfree 128
> >
> > Also i post 2 vmstat outputs one with normal workload and one with the
> > performance problems.
> >
> > NORMAL
> >
> >
> > kthr memory page faults cpu
> > ----- ----------- ------------------------ ------------ -----------
> > r b avm fre re pi po fr sr cy in sy cs us sy id wa
> > 2 2 178921 121 0 4 3 317 290 0 194 318 222 22 9 60 9
> > 2 2 178921 128 0 12 5 113 159 0 819 6716 2565 63 16 13 7
> > 2 2 178921 126 0 3 3 89 122 0 797 5971 2414 60 14 19 7
> > 1 2 178921 123 0 2 1 50 75 0 477 2566 993 36 4 53 7
> > 1 2 178921 125 0 2 3 73 120 0 549 2662 1243 45 5 40 10
> >
> > PROBLEM
> >
> > kthr memory page faults cpu
> > ----- ----------- ------------------------ ------------ -----------
> > r b avm fre re pi po fr sr cy in sy cs us sy id wa
> > 2 2 178979 123 0 4 3 317 290 0 194 322 224 22 9 60 9
> > 2 2 178979 124 0 12 27 792 1973 0 1071 4877 3069 63 15 4 18
> > 4 3 179532 112 0 66 37 889 1561 0 1150 3573 4763 72 17 2 9
> > 3 3 179532 119 0 13 13 728 1319 0 1072 3792 4546 73 16 2 9
> > 4 3 179532 124 0 9 65 733 1487 0 1135 4299 4041 66 21 3 10
> > 5 2 179812 92 0 14 72 891 1666 0 1175 4183 3866 73 15 2 10
> > 5 3 179881 123 0 10 156 851 3283 0 1153 2445 2031 36 8 10 45
> > 2 4 179896 0 0 15 182 947 2160 0 1250 2719 2428 31 10 7 52
> > 1 5 180565 32 0 7 189 596 2143 0 966 1563 1478 14 6 8 72
> > 3 4 180655 107 0 53 104 688 1749 0 1148 4207 3253 60 13 2 25
> > 3 4 179879 906 0 17 129 625 1167 0 1078 2961 3994 56 13 3 28
> > 1 5 180201 0 0 50 107 765 1541 0 1102 2609 2824 29 9 8 54
> > 1 4 181176 94 0 19 167 961 2153 0 1106 2439 2070 26 9 5 60
> > 6 5 181338 117 0 87 59 633 1408 0 1138 6255 3419 63 14 2 21
> > 2 4 181338 0 0 26 129 639 1952 0 1009 2332 1875 38 7 11 44
> > 3 2 180957 125 0 50 130 403 940 0 915 2771 1677 33 7 9 51
> >
> > I have determinate that we use about 1.4 GB for file caching. I want
> > to try reduce minperm and maxperm in order to allow at max about 30%
> > of RAM for file caching and i think i should gain 700MB free memory
> > for oracle.
> >
> > Is my thought correct ?
> >
> > Has someone faced the same problem ?
> >
> > Is anything else i can do ?
> >
> >
> >
> > Thanks in abvance and sorry for the extened thread.
> >
> > Agis
> >
> > ORACLE DBA

>
>
> I would start with minperm = 10 & maxperm = 30, that will definitely
> help your performance.
>
> What version of AIX?
>
> Do you have Asynch I/O enabled? Are you mounting the filesystem with
> the Oracle tablespaces with Direct I/O ?
>
> Ralph

AIX v 4.3.3

Asynch I/O enabled

Filesystem is JFS.

So i think i can go through your advices and my assuption to use vmtune -p10 -P30

[Jurjen Oskam]

In article <3F66BC35.3080407@vankoll.de>, Holger van Koll wrote:

> That only works if oracle runs as root, doesnt it?
> At least I see no way for a non-root task to pin memory.
>
> Btw. its obsoleted
>
> [url]http://www.orafaq.com/parms/parm369.htm[/url]

Ah, well, scratch that one then. My excuse is that I'm not an Oracle
DBA. :-)

Thanks for the information.

--
Jurjen Oskam
(sorry for messing up the threading - posting via Google again, the article
I'm following up to isn't yet known to Google)

PGP Key available at [url]http://www.stupendous.org/[/url]

[agis]

[email]agstamy@yahoo.com[/email] (agis) wrote in message news:<d2aa3279.0309152341.dc92ee0@posting.google.c om>...

> [email]ralph.rye@norgetek.com[/email] (Ralph) wrote in message news:<57422359.0309151205.51e11360@posting.google. com>...

> > [email]agstamy@yahoo.com[/email] (agis) wrote in message news:<d2aa3279.0309150615.4c673a1@posting.google.c om>...

> > > Hi to all,
> > >
> > > I have an RS/6000 with2 CPUs and 2048 MB of RAM running tow instances
> > > of ORACLE 8.1.7 database.One with 440MB SGA and with about 20
> > > dedicated connections and about 100 in shared servers and one with
> > > 35MB SGA with almost no activity on JFS file system.
> > >
> > > Some times on peak hours we face performance problems in response
> > > time. Oracle is tuned and with vmstat i realized that we had a lot of
> > > paging and swapping. I use nmon utility with the folwoing output :
> > >
> > > % Used 100.0% 31.1% to Paging Space 12.0 29.0
> > > numperm 73.1%
> > > % Free 0.0% 68.9% to File System 457.6 815.3
> > > minperm 20.0%
> > > MB Used 2047.5MB 636.2MB Page Scans 407.7
> > > maxperm 79.8%
> > > MB Free 0.5MB 1411.8MB Page Cycles 0.0
> > > minfree 120
> > > Total(MB) 2048.0MB 2048.0MB Page Reclaim 0.0
> > > maxfree 128
> > >
> > > Also i post 2 vmstat outputs one with normal workload and one with the
> > > performance problems.
> > >
> > > NORMAL
> > >
> > >
> > > kthr memory page faults cpu
> > > ----- ----------- ------------------------ ------------ -----------
> > > r b avm fre re pi po fr sr cy in sy cs us sy id wa
> > > 2 2 178921 121 0 4 3 317 290 0 194 318 222 22 9 60 9
> > > 2 2 178921 128 0 12 5 113 159 0 819 6716 2565 63 16 13 7
> > > 2 2 178921 126 0 3 3 89 122 0 797 5971 2414 60 14 19 7
> > > 1 2 178921 123 0 2 1 50 75 0 477 2566 993 36 4 53 7
> > > 1 2 178921 125 0 2 3 73 120 0 549 2662 1243 45 5 40 10
> > >
> > > PROBLEM
> > >
> > > kthr memory page faults cpu
> > > ----- ----------- ------------------------ ------------ -----------
> > > r b avm fre re pi po fr sr cy in sy cs us sy id wa
> > > 2 2 178979 123 0 4 3 317 290 0 194 322 224 22 9 60 9
> > > 2 2 178979 124 0 12 27 792 1973 0 1071 4877 3069 63 15 4 18
> > > 4 3 179532 112 0 66 37 889 1561 0 1150 3573 4763 72 17 2 9
> > > 3 3 179532 119 0 13 13 728 1319 0 1072 3792 4546 73 16 2 9
> > > 4 3 179532 124 0 9 65 733 1487 0 1135 4299 4041 66 21 3 10
> > > 5 2 179812 92 0 14 72 891 1666 0 1175 4183 3866 73 15 2 10
> > > 5 3 179881 123 0 10 156 851 3283 0 1153 2445 2031 36 8 10 45
> > > 2 4 179896 0 0 15 182 947 2160 0 1250 2719 2428 31 10 7 52
> > > 1 5 180565 32 0 7 189 596 2143 0 966 1563 1478 14 6 8 72
> > > 3 4 180655 107 0 53 104 688 1749 0 1148 4207 3253 60 13 2 25
> > > 3 4 179879 906 0 17 129 625 1167 0 1078 2961 3994 56 13 3 28
> > > 1 5 180201 0 0 50 107 765 1541 0 1102 2609 2824 29 9 8 54
> > > 1 4 181176 94 0 19 167 961 2153 0 1106 2439 2070 26 9 5 60
> > > 6 5 181338 117 0 87 59 633 1408 0 1138 6255 3419 63 14 2 21
> > > 2 4 181338 0 0 26 129 639 1952 0 1009 2332 1875 38 7 11 44
> > > 3 2 180957 125 0 50 130 403 940 0 915 2771 1677 33 7 9 51
> > >
> > > I have determinate that we use about 1.4 GB for file caching. I want
> > > to try reduce minperm and maxperm in order to allow at max about 30%
> > > of RAM for file caching and i think i should gain 700MB free memory
> > > for oracle.
> > >
> > > Is my thought correct ?
> > >
> > > Has someone faced the same problem ?
> > >
> > > Is anything else i can do ?
> > >
> > >
> > >
> > > Thanks in abvance and sorry for the extened thread.
> > >
> > > Agis
> > >
> > > ORACLE DBA

> >
> >
> > I would start with minperm = 10 & maxperm = 30, that will definitely
> > help your performance.
> >
> > What version of AIX?
> >
> > Do you have Asynch I/O enabled? Are you mounting the filesystem with
> > the Oracle tablespaces with Direct I/O ?
> >
> > Ralph

>
>
> AIX v 4.3.3
>
> Asynch I/O enabled
>
> Filesystem is JFS.
>
> So i think i can go through your advices and my assuption to use vmtune -p10 -P30

LOCK_SGA is suported in AIX ver 4.3.3 and up and for ORACLE 8i and up.
I will try first the AIX tuning and the Oracle SGA fitted in memory.
Any other suggestions or i should proceed?

[Mike Marshall]

[email]agstamy@yahoo.com[/email] (agis) writes:

>I will try first the AIX tuning and the Oracle SGA fitted in memory.
>Any other suggestions or i should proceed?

Good luck with tuning, I hope that works out. I'm not an AIX (tm)
sysadmin, as such, but my rule of thumb on Solaris is that if
vmstat's scan-rate (sr) column isn't nearly zero most of the time, then
you don't have enough memory to do whatever it is you're trying
to do. The system is thrashing around hunting for pages to use.

-Mike

[agis]

[email]hubcap@hubcap.clemson.edu[/email] (Mike Marshall) wrote in message news:<bkd0kq$jcg$1@hubcap.clemson.edu>...

> [email]agstamy@yahoo.com[/email] (agis) writes:

> >I will try first the AIX tuning and the Oracle SGA fitted in memory.
> >Any other suggestions or i should proceed?

>
> Good luck with tuning, I hope that works out. I'm not an AIX (tm)
> sysadmin, as such, but my rule of thumb on Solaris is that if
> vmstat's scan-rate (sr) column isn't nearly zero most of the time, then
> you don't have enough memory to do whatever it is you're trying
> to do. The system is thrashing around hunting for pages to use.
>
> -Mike

Well , i issued vmtune -p10 -P30 But numperm was steal to 73%.

Then i issued vmtune -h1 and numperm falled immediately to 30%.

Then i issued vmtune -h0 and numeprm slowly increases and becomes
greater than maxperm. Why this haapen? (I think i have no need for
much memory so AIX gives memory to JFS.)

But can i leave strict_maxperm to 1, because i read in some posts that
it is not working as it is documented !!!!

I am waiting for further oppinions.

Thanks AGIS

[Ralph]

[email]agstamy@yahoo.com[/email] (agis) wrote in message news:<d2aa3279.0309230148.5a67cc3c@posting.google. com>...

> [email]hubcap@hubcap.clemson.edu[/email] (Mike Marshall) wrote in message news:<bkd0kq$jcg$1@hubcap.clemson.edu>...

> > [email]agstamy@yahoo.com[/email] (agis) writes:

> > >I will try first the AIX tuning and the Oracle SGA fitted in memory.
> > >Any other suggestions or i should proceed?

> >
> > Good luck with tuning, I hope that works out. I'm not an AIX (tm)
> > sysadmin, as such, but my rule of thumb on Solaris is that if
> > vmstat's scan-rate (sr) column isn't nearly zero most of the time, then
> > you don't have enough memory to do whatever it is you're trying
> > to do. The system is thrashing around hunting for pages to use.
> >
> > -Mike

>
>
> Well , i issued vmtune -p10 -P30 But numperm was steal to 73%.
>
> Then i issued vmtune -h1 and numperm falled immediately to 30%.
>
> Then i issued vmtune -h0 and numeprm slowly increases and becomes
> greater than maxperm. Why this haapen? (I think i have no need for
> much memory so AIX gives memory to JFS.)
>
> But can i leave strict_maxperm to 1, because i read in some posts that
> it is not working as it is documented !!!!
>
> I am waiting for further oppinions.
>
> Thanks AGIS

Agis,

If you mount your Oracle data filesystems direct I/O, Oracle will then
bypass the OS filesystem cache altogether. So then you should not see
your numperm rising above maxperm, if the only process doing large
amounts of I/O are Oracle on this system.

There are pros/cons about using direct I/O, depending on the workload
of your database you could see great benefits from it.

Ralph

[mark taylor]

> > > Good luck with tuning, I hope that works out. I'm not an AIX (tm)

> > > sysadmin, as such, but my rule of thumb on Solaris is that if
> > > vmstat's scan-rate (sr) column isn't nearly zero most of the time, then
> > > you don't have enough memory to do whatever it is you're trying
> > > to do. The system is thrashing around hunting for pages to use.

Solaris uses a completely different page stealing algorythm to AIX, so
you cannot compare the 2.

AIX uses LRUD (least recently used) and Solaris uses LIFO (last in 1st
out).

>> scan-rate (sr) column isn't nearly zero most of the time.

Again, AIX will build as large a filesystem cache as possible by
default. when it hits minperm it is going to scan for pages to free up
and free them according to LRUD algorythm... the pages it frees up are
dependant on the number of filepages cached and the maxperm / minperm
settings.

If numperm is above maxperm it is non discriminate over what pages to
mark as candidates to free up, but if numperm is below maxperm, then
it will only mark file (persistient) pages as candidates to get the
size of the fs cache down.

NOTE: by default it only does this once you hit minfree..

To strictly set the maximum number of file pages cached you would set
strict_maxperm, but you usually do not have to do this unless you are
working with a very large amount of memory (64Gb and up) ... so, i
would leave well alone if you only have a couple of GB...


Ok, Now for tuning... you use JFS DB files ? correct ? in which case
there is a lot of benefit in AIX caching pages from you DB files...
BUT, and this is a BIG BUT.. You don't want to do this at the
detrement of you SGA.. SGA must always remain in realmem due to the
nature of working (computational) storage mem pages, you are highly
likely to nee that page again soom.. this is how thrashing happens..
paging out to paging space only to have to page back in..

lets do the Maths..

taking the "avm" from vmstat 178921 (4k pages)
178921*4096
732860416/1024/1024 = 698MB active virtual memory ...

698/2048*100=34%

So, you can set maxperm to 65% if you require ... but, you need to
allow some lea way..

30% and 15% were recommeded further up the thread. I would go with
that..

then you need to monitor for pi/po.. if you still have some, then
paste the output to

vmstat 1 15
/usr/samples/kernel/vmtune (and -a)

Rgds
Mark

[agis]

[email]mky@talk21.com[/email] (mark taylor) wrote in message news:<dee0a7c3.0309250417.2f1ef77d@posting.google. com>...

> > > > Good luck with tuning, I hope that works out. I'm not an AIX (tm)
> > > > sysadmin, as such, but my rule of thumb on Solaris is that if
> > > > vmstat's scan-rate (sr) column isn't nearly zero most of the time, then
> > > > you don't have enough memory to do whatever it is you're trying
> > > > to do. The system is thrashing around hunting for pages to use.

>
> Solaris uses a completely different page stealing algorythm to AIX, so
> you cannot compare the 2.
>
> AIX uses LRUD (least recently used) and Solaris uses LIFO (last in 1st
> out).
>

> >> scan-rate (sr) column isn't nearly zero most of the time.

>
> Again, AIX will build as large a filesystem cache as possible by
> default. when it hits minperm it is going to scan for pages to free up
> and free them according to LRUD algorythm... the pages it frees up are
> dependant on the number of filepages cached and the maxperm / minperm
> settings.
>
> If numperm is above maxperm it is non discriminate over what pages to
> mark as candidates to free up, but if numperm is below maxperm, then
> it will only mark file (persistient) pages as candidates to get the
> size of the fs cache down.
>
> NOTE: by default it only does this once you hit minfree..
>
> To strictly set the maximum number of file pages cached you would set
> strict_maxperm, but you usually do not have to do this unless you are
> working with a very large amount of memory (64Gb and up) ... so, i
> would leave well alone if you only have a couple of GB...
>
>
> Ok, Now for tuning... you use JFS DB files ? correct ? in which case
> there is a lot of benefit in AIX caching pages from you DB files...
> BUT, and this is a BIG BUT.. You don't want to do this at the
> detrement of you SGA.. SGA must always remain in realmem due to the
> nature of working (computational) storage mem pages, you are highly
> likely to nee that page again soom.. this is how thrashing happens..
> paging out to paging space only to have to page back in..
>
> lets do the Maths..
>
> taking the "avm" from vmstat 178921 (4k pages)
> 178921*4096
> 732860416/1024/1024 = 698MB active virtual memory ...
>
> 698/2048*100=34%
>
> So, you can set maxperm to 65% if you require ... but, you need to
> allow some lea way..
>
> 30% and 15% were recommeded further up the thread. I would go with
> that..
>
> then you need to monitor for pi/po.. if you still have some, then
> paste the output to
>
> vmstat 1 15
> /usr/samples/kernel/vmtune (and -a)
>
> Rgds
> Mark

Thanks for the advice.

I use ORACLE on JFS.

I have set minperm=10 and maxperm=30 and strict_maxperm deactivated.

pi/po on vmstat output now is almost zero(very few times some minor
activity of pi/po)

I also set vmtune -W32 -c2 -s1
This also sets the maxrandwrt3 to 32*4 KB and numclust4 to 2*16 KB and
activates the sync_release_ilock parameter.

according the IBM RS/6000 SP System Performance Tuning Update

Performance seems more stabilized but there area crises momements that
i have the behavior,like before tuning, but with no pi/po activity
!!!!

Do you think i should set maxpgahead to 16 (is default 8) and set
acordingly the minfree maxfree ?

Thanks

Agis