Very Computer

by **Norman Levi** » Thu, 12 Dec 2002 07:59:28

I am trying to find 'definitive' words about AIX paging, but for now, I will eat
crow (in garlic sauce) and give you some info from a friend of mine who has been
in AIX for a good number of years and used to teach VM internals with me - so I
trust his knowledge.
For some reason, he entered his data in upper case - so don't yell at me....

does aix do anything smart like writing pages to area with shortest i/o
queue? or is it really as simple minded as documentation would
indicate - that is a round robin approach.

ROUND ROBIN AS FAR AS I KNOW. PAGING IN GENERAL IS DISCOURAGED.
RECOMMENDATION IS TO BUY ENOUGH RAM TO HOLD YOUR APPS, APP WORKING
STORAGE AND FILE I/O SINCE FILE I/O IS HANDLED BY THE PAGING SUBSYSTEM.
SINCE RAM IS CHEAP TODAY THIS IS EVEN MORE THE REASON TO GO WITH THE
RECOMMENDATION. STOP PAGING, BUY RAM. UNIX PAGING WAS DESIGNED AS A
SAFETY DEVICE. IT WAS NEVER INTENDED TO BE USED. THIS EXPLAINS WHY NO
EFFORT WAS EVER PUT INTO MAKING IT EFFICIENT. THERE IS ALSO NOTHING
ANALOGOUS TO MVS SWAPPING.

Any thoughts on validity of equal size areas for page? I think it is
garbage.

ALLOCATION IS ROUND ROBIN UNTIL THE SMALLER AREA FILLS UP. THE ONLY REAL
BENEFIT IS THAT YOU SPREAD THE PAGING I/O ACROSS MORE THAN ONE DEVICE.

--
Norman Levin - to send a NEW note to me, click on the following line:
mailto:%6E%6F%72%6D%61%6E%6C%65%76%69%6E%40%61%74%74%67%6C%6F%62%61%6C%2E%6E%65%7
4

by **bill** » Thu, 12 Dec 2002 12:40:36

> I am trying to find 'definitive' words about AIX paging, but for now, I will eat
> crow (in garlic sauce) and give you some info from a friend of mine who has been
> in AIX for a good number of years and used to teach VM internals with me - so I
> trust his knowledge.
> For some reason, he entered his data in upper case - so don't yell at me....

> does aix do anything smart like writing pages to area with shortest i/o
> queue? or is it really as simple minded as documentation would
> indicate - that is a round robin approach.

> ROUND ROBIN AS FAR AS I KNOW. PAGING IN GENERAL IS DISCOURAGED.
> RECOMMENDATION IS TO BUY ENOUGH RAM TO HOLD YOUR APPS, APP WORKING
> STORAGE AND FILE I/O SINCE FILE I/O IS HANDLED BY THE PAGING SUBSYSTEM.
> SINCE RAM IS CHEAP TODAY THIS IS EVEN MORE THE REASON TO GO WITH THE
> RECOMMENDATION. STOP PAGING, BUY RAM. UNIX PAGING WAS DESIGNED AS A
> SAFETY DEVICE. IT WAS NEVER INTENDED TO BE USED. THIS EXPLAINS WHY NO
> EFFORT WAS EVER PUT INTO MAKING IT EFFICIENT. THERE IS ALSO NOTHING
> ANALOGOUS TO MVS SWAPPING.

> Any thoughts on validity of equal size areas for page? I think it is
> garbage.

> ALLOCATION IS ROUND ROBIN UNTIL THE SMALLER AREA FILLS UP. THE ONLY REAL
> BENEFIT IS THAT YOU SPREAD THE PAGING I/O ACROSS MORE THAN ONE DEVICE.

> --
> Norman Levin - to send a NEW note to me, click on the following line:
> mailto:%6E%6F%72%6D%61%6E%6C%65%76%69%6E%40%61%74%74%67%6C%6F%62%61%6C%2E%6E%65%7
> 4

ah real computers mvs ..... your friend is right. paging in a unix box should be
rare.... unix in a risc environment is for speed... a lot of old flat file
apps.. database apps needs retuning of disk io not more memory.. if a system is
paging buy more ram... the real bottle neck in most systems today appears to be
133meg bus.... (i still have some MCA machines).
ah channels

by **Simon Marches** » Thu, 12 Dec 2002 21:45:38

>>I am trying to find 'definitive' words about AIX paging, but for now, I will eat
>>crow (in garlic sauce) and give you some info from a friend of mine who has been
>>in AIX for a good number of years and used to teach VM internals with me - so I
>>trust his knowledge.
>>For some reason, he entered his data in upper case - so don't yell at me....

>>does aix do anything smart like writing pages to area with shortest i/o
>>queue? or is it really as simple minded as documentation would
>>indicate - that is a round robin approach.

>>ROUND ROBIN AS FAR AS I KNOW. PAGING IN GENERAL IS DISCOURAGED.
>>RECOMMENDATION IS TO BUY ENOUGH RAM TO HOLD YOUR APPS, APP WORKING
>>STORAGE AND FILE I/O SINCE FILE I/O IS HANDLED BY THE PAGING SUBSYSTEM.
>>SINCE RAM IS CHEAP TODAY THIS IS EVEN MORE THE REASON TO GO WITH THE
>>RECOMMENDATION. STOP PAGING, BUY RAM. UNIX PAGING WAS DESIGNED AS A
>>SAFETY DEVICE. IT WAS NEVER INTENDED TO BE USED. THIS EXPLAINS WHY NO
>>EFFORT WAS EVER PUT INTO MAKING IT EFFICIENT. THERE IS ALSO NOTHING
>>ANALOGOUS TO MVS SWAPPING.

>>Any thoughts on validity of equal size areas for page? I think it is
>>garbage.

>>ALLOCATION IS ROUND ROBIN UNTIL THE SMALLER AREA FILLS UP. THE ONLY REAL
>>BENEFIT IS THAT YOU SPREAD THE PAGING I/O ACROSS MORE THAN ONE DEVICE.

>>--
>>Norman Levin - to send a NEW note to me, click on the following line:
>>mailto:%6E%6F%72%6D%61%6E%6C%65%76%69%6E%40%61%74%74%67%6C%6F%62%61%6C%2E%6E%65%7
>>4

> ah real computers mvs ..... your friend is right. paging in a unix box should be
> rare.... unix in a risc environment is for speed... a lot of old flat file
> apps.. database apps needs retuning of disk io not more memory.. if a system is
> paging buy more ram... the real bottle neck in most systems today appears to be
> 133meg bus.... (i still have some MCA machines).
> ah channels

For a reason which now escapes me, I have been doing some reading about
this recently. AIX (including v1/v2 for the RT/6150) uses an inverted
page table as described in the "RISC System/6000 Technology Guide". This
means that there are cases when a page fault will actually incur two
sequential page faults - one for the relevant external page table entry
and one for the page itself. According to Operating System Concepts 3rd
Edition (Silbershatz/Peterson/Galvin) pp316-7, the use of inverted page
tables is a trade off to reduce the amount of memory used in page tables
at the cost of slower page fault resolution if paging is actually
required. This would be particularly relevant to AIX because of the very
large address space used to implement AIX's version of single level
storage which has some similarities to that in the S/38 AS/400 iSeries
systems.

by **Holger van Kol** » Fri, 13 Dec 2002 00:07:36

> this recently. AIX (including v1/v2 for the RT/6150) uses an inverted
> page table as described in the "RISC System/6000 Technology Guide". This
> means that there are cases when a page fault will actually incur two
> sequential page faults - one for the relevant external page table entry
> and one for the page itself.

Yes. vmstat -s will tell you (backtracks) if this happens

by **Timothy J. Boga** » Fri, 13 Dec 2002 02:14:11

>I am trying to find 'definitive' words about AIX paging, but for now, I will eat
>crow (in garlic sauce) and give you some info from a friend of mine who has been
>in AIX for a good number of years and used to teach VM internals with me - so I
>trust his knowledge.
>For some reason, he entered his data in upper case - so don't yell at me....

<snip>

Ah, a taste I know well. 8-) We are all trying to find the real answer,
whether it is the one we want or not - eh?

Sort of reminds me of the CE I had in Houston who would come in to install something,
sit down at the console, hit the CAPS LOCK and start typing the commands from the
installation instructions....

8-)

--
"I'm a big fan of ignorance based techniques, because humans
have a lot of ignorance, and we want to play our strong suit."
Eric Lander New Scientist

by **Douglas R. Probs** » Sat, 14 Dec 2002 12:29:27

Ok here is what I do.
If you are running 4.3.3 or lower real mem x1 for paging. this is cheap ..
disk is cheaper then RAM.
if you are at 5.1 or higher real mem x1 .. it will never use any of it if
you have enough RAM for your app but disk is cheap.. be safe.. be available
..sleep well.
Doug

Quote:> I am trying to find 'definitive' words about AIX paging, but for now, I
will eat
> crow (in garlic sauce) and give you some info from a friend of mine who
has been
> in AIX for a good number of years and used to teach VM internals with me -
so I
> trust his knowledge.
> For some reason, he entered his data in upper case - so don't yell at
me....

> does aix do anything smart like writing pages to area with shortest i/o
> queue? or is it really as simple minded as documentation would
> indicate - that is a round robin approach.

> ROUND ROBIN AS FAR AS I KNOW. PAGING IN GENERAL IS DISCOURAGED.
> RECOMMENDATION IS TO BUY ENOUGH RAM TO HOLD YOUR APPS, APP WORKING
> STORAGE AND FILE I/O SINCE FILE I/O IS HANDLED BY THE PAGING SUBSYSTEM.
> SINCE RAM IS CHEAP TODAY THIS IS EVEN MORE THE REASON TO GO WITH THE
> RECOMMENDATION. STOP PAGING, BUY RAM. UNIX PAGING WAS DESIGNED AS A
> SAFETY DEVICE. IT WAS NEVER INTENDED TO BE USED. THIS EXPLAINS WHY NO
> EFFORT WAS EVER PUT INTO MAKING IT EFFICIENT. THERE IS ALSO NOTHING
> ANALOGOUS TO MVS SWAPPING.

> Any thoughts on validity of equal size areas for page? I think it is
> garbage.

> ALLOCATION IS ROUND ROBIN UNTIL THE SMALLER AREA FILLS UP. THE ONLY REAL
> BENEFIT IS THAT YOU SPREAD THE PAGING I/O ACROSS MORE THAN ONE DEVICE.

> --
> Norman Levin - to send a NEW note to me, click on the following line:

mailto:%6E%6F%72%6D%61%6E%6C%65%76%69%6E%40%61%74%74%67%6C%6F%62%61%6C%2E%6E
%65%7

- Show quoted text -

Quote:> 4

by **Edward Melto** » Wed, 18 Dec 2002 17:07:30

Starting at v4.3.3, I believe, IBM removed the requirement for page
space (read doesn't automatically use it). I've also always tried to do
at least 1X RAM for a page space but have started to do 512-1GB for
systems with 4GB and higher. Take note though, if you do this then keep
a VERY close eye on paging space utilization and make sure your apps
aren't using it (SAP/Oracle are notorious for this). Douglas is correct
that disk is cheaper than RAM, in fact, MUCH cheaper. However, a poorly
performing server(s) could cost you your job. My advice is the same as
Norman's -- don't rely on paging space! Keep an eye on vmstat from time
to time and if you start seeing PI's and PO's put a request in for
additional RAM (e-mail for documentation). THAT is doing what a system's
engineer is supposed to do. If the powers that be put the kabosh on
spending more $$$ for RAM then you can forward user complaints to the
kabosher until he agrees to buy more memory! :)

Ed

> Ok here is what I do.
> If you are running 4.3.3 or lower real mem x1 for paging. this is cheap ..
> disk is cheaper then RAM.
> if you are at 5.1 or higher real mem x1 .. it will never use any of it if
> you have enough RAM for your app but disk is cheap.. be safe.. be available
> ..sleep well.
> Doug

>>I am trying to find 'definitive' words about AIX paging, but for now, I

> will eat

>>crow (in garlic sauce) and give you some info from a friend of mine who

> has been

>>in AIX for a good number of years and used to teach VM internals with me -

> so I

>>trust his knowledge.
>>For some reason, he entered his data in upper case - so don't yell at

> me....

>>does aix do anything smart like writing pages to area with shortest i/o
>>queue? or is it really as simple minded as documentation would
>>indicate - that is a round robin approach.

>>ROUND ROBIN AS FAR AS I KNOW. PAGING IN GENERAL IS DISCOURAGED.
>>RECOMMENDATION IS TO BUY ENOUGH RAM TO HOLD YOUR APPS, APP WORKING
>>STORAGE AND FILE I/O SINCE FILE I/O IS HANDLED BY THE PAGING SUBSYSTEM.
>>SINCE RAM IS CHEAP TODAY THIS IS EVEN MORE THE REASON TO GO WITH THE
>>RECOMMENDATION. STOP PAGING, BUY RAM. UNIX PAGING WAS DESIGNED AS A
>>SAFETY DEVICE. IT WAS NEVER INTENDED TO BE USED. THIS EXPLAINS WHY NO
>>EFFORT WAS EVER PUT INTO MAKING IT EFFICIENT. THERE IS ALSO NOTHING
>>ANALOGOUS TO MVS SWAPPING.

>>Any thoughts on validity of equal size areas for page? I think it is
>>garbage.

>>ALLOCATION IS ROUND ROBIN UNTIL THE SMALLER AREA FILLS UP. THE ONLY REAL
>>BENEFIT IS THAT YOU SPREAD THE PAGING I/O ACROSS MORE THAN ONE DEVICE.

>>--
>>Norman Levin - to send a NEW note to me, click on the following line:

> mailto:%6E%6F%72%6D%61%6E%6C%65%76%69%6E%40%61%74%74%67%6C%6F%62%61%6C%2E%6E
> %65%7

>>4

Very Computer

AIX Paging - maybe eating crow...

AIX Paging - maybe eating crow...

AIX Paging - maybe eating crow...

AIX Paging - maybe eating crow...

AIX Paging - maybe eating crow...

AIX Paging - maybe eating crow...

AIX Paging - maybe eating crow...

AIX Paging - maybe eating crow...