AIX concurrent IO == Linux Asynchronous I/O

[Troels Arvin]

Hello,

Running DB2 on AIX.

I've read
http://www-128.ibm.com/developerwork...le/dm-0408lee/

DMS tablespaces on concurrent I/O file systems (with the AUTORESIZE YES
option, for manageability) sounds like it's exactly the thing for me.

However, I'm /truly/ annoyed by AIX. Today at a trade show, an IBM
consultant demonstrated how Linux can be run on pSeries systems LPARs,
just as fine as AIX. The thought of moving to DB2 on Linux on PowerPC is
intriguing.

But: Is it possible for DB2 to work with concurrent I/O on Linux? In some
DB2 papers, I've read about DB2 "asynchoronous I/O" on Linux; is that
equivalent to concurrent IO on AIX?

--
Regards,
Troels Arvin <tr****@arvin.dk>
http://troels.arvin.dk/

[Haider Rizvi]

No, asynch i/o on Linux is different from cio/dio on AIX. On both
linux and aix, direct i/o (i.e., avoiding filesystem caching) is used
/ attained by the 'no filesystem caching' clause on a tablespace
definition.

Asynchronous i/o is the ability to issue i/os and not having to
synchronously wait for the i/o to complete (hence the
name). DB2Linuxaio is the registry variable to use for that.

Troels Arvin <tr****@arvin.dkwrites:

Hello,

Running DB2 on AIX.

I've read
http://www-128.ibm.com/developerwork...le/dm-0408lee/

DMS tablespaces on concurrent I/O file systems (with the
AUTORESIZE YES option, for manageability) sounds like it's exactly
the thing for me.

However, I'm /truly/ annoyed by AIX. Today at a trade show, an IBM
consultant demonstrated how Linux can be run on pSeries systems
LPARs, just as fine as AIX. The thought of moving to DB2 on Linux
on PowerPC is intriguing.

But: Is it possible for DB2 to work with concurrent I/O on Linux?
In some DB2 papers, I've read about DB2 "asynchoronous I/O" on
Linux; is that equivalent to concurrent IO on AIX?

--
Regards,
Troels Arvin <tr****@arvin.dk>
http://troels.arvin.dk/

--
Haider

[Philip Nelson]

And further to this, I recall that Linux AIO is now (V9) on by default, and
the registry variable is deprecated.

Phil

Haider Rizvi wrote:

No, asynch i/o on Linux is different from cio/dio on AIX. On both
linux and aix, direct i/o (i.e., avoiding filesystem caching) is used
/ attained by the 'no filesystem caching' clause on a tablespace
definition.

Asynchronous i/o is the ability to issue i/os and not having to
synchronously wait for the i/o to complete (hence the
name). DB2Linuxaio is the registry variable to use for that.

Troels Arvin <tr****@arvin.dkwrites:

>Hello,

Running DB2 on AIX.

I've read
http://www-128.ibm.com/developerwork...le/dm-0408lee/

DMS tablespaces on concurrent I/O file systems (with the
AUTORESIZE YES option, for manageability) sounds like it's exactly
the thing for me.

However, I'm /truly/ annoyed by AIX. Today at a trade show, an IBM
consultant demonstrated how Linux can be run on pSeries systems
LPARs, just as fine as AIX. The thought of moving to DB2 on Linux
on PowerPC is intriguing.

But: Is it possible for DB2 to work with concurrent I/O on Linux?
In some DB2 papers, I've read about DB2 "asynchoronous I/O" on
Linux; is that equivalent to concurrent IO on AIX?

--
Regards,
Troels Arvin <tr****@arvin.dk>
http://troels.arvin.dk/

[Troels Arvin]

Hello,

On Fri, 02 Mar 2007 19:03:17 -0500, Haider Rizvi wrote:

No, asynch i/o on Linux is different from cio/dio on AIX.

OK :-(

I read http://doi.acm.org/10.1145/1146847.1146890
I seems that with DB2 for Linux, one has to accept a 5% performance hit
if tablespaces aren't stored on raw devices. - While http://
www-128.ibm.com/developerworks/db2/library/techarticle/dm-0408lee/ seems
to indicate that on AIX you can have non-raw-device-storage without
performance degration.

--
Regards,
Troels Arvin <tr****@arvin.dk>
http://troels.arvin.dk/

[Haider Rizvi]

Hi Troels,

As always performance data is a 'it depends'. In either paper the
point is that with the direct i/o approach for various OSes the
filesystems performance is approaching that of raw devices. And the
administration of filesystems is generally considered to be easier in
a dynamic environment (I know some would disagree with even that
statement).

That's why using filesystems-based dms containers and using direct i/o
(i.e., "no filesystem caching" clause) is our best practices
recommendation at this point, see for example the BCU documentation.

Regards,
--
Haider
Troels Arvin <tr****@arvin.dkwrites:

Hello,

On Fri, 02 Mar 2007 19:03:17 -0500, Haider Rizvi wrote:

>No, asynch i/o on Linux is different from cio/dio on AIX.

OK :-(

I read http://doi.acm.org/10.1145/1146847.1146890 I seems that
with DB2 for Linux, one has to accept a 5% performance hit if
tablespaces aren't stored on raw devices. - While http://
www-128.ibm.com/developerworks/db2/library/techarticle/dm-0408lee/
seems to indicate that on AIX you can have non-raw-device-storage
without performance degration.

--
Regards,
Troels Arvin <tr****@arvin.dk>
http://troels.arvin.dk/

--
Haider

[Mark A]

"Troels Arvin" <tr****@arvin.dkwrote in message
news:pa*********************@arvin.dk...

I read http://doi.acm.org/10.1145/1146847.1146890
I seems that with DB2 for Linux, one has to accept a 5% performance hit
if tablespaces aren't stored on raw devices. - While http://
www-128.ibm.com/developerworks/db2/library/techarticle/dm-0408lee/ seems
to indicate that on AIX you can have non-raw-device-storage without
performance degration.

It depends on a number of factors:

1. Is it a data warehouse or OLTP application? If it is a data warehouse AND
most of the queries will be doing table scans (reading the entire table) on
large tables, then raw devices provide slightly better optimization of the
table and tablespace placement for such queries. If the application is not
doing table scans on large tables, you will not notice any difference.

2. Bufferpool hit ratio. If the database has a high bufferpool hit ratio,
then there is not much synchronous I/O, and raw devices will provide very
(or no) measurable benefit.

3. The type of disk you are using. If you use a SAN or NAS and have no
control over where the containers are actually placed, then raw devices will
have little or no benefit.

[Haider Rizvi]

The db2linuxaio registry variable should continue to be used with V9
where it is appropriate.

- Haider

Philip Nelson <te*****@scotdb.comwrites:

And further to this, I recall that Linux AIO is now (V9) on by
default, and the registry variable is deprecated.

Phil

Haider Rizvi wrote:

>No, asynch i/o on Linux is different from cio/dio on AIX. On
both linux and aix, direct i/o (i.e., avoiding filesystem
caching) is used / attained by the 'no filesystem caching'
clause on a tablespace definition.

Asynchronous i/o is the ability to issue i/os and not having to
synchronously wait for the i/o to complete (hence the
name). DB2Linuxaio is the registry variable to use for that.

Troels Arvin <tr****@arvin.dkwrites:

>>Hello,

Running DB2 on AIX.

I've read
http://www-128.ibm.com/developerwork...le/dm-0408lee/

DMS tablespaces on concurrent I/O file systems (with the
AUTORESIZE YES option, for manageability) sounds like it's
exactly the thing for me.

However, I'm /truly/ annoyed by AIX. Today at a trade show, an
IBM consultant demonstrated how Linux can be run on pSeries
systems LPARs, just as fine as AIX. The thought of moving to
DB2 on Linux on PowerPC is intriguing.

But: Is it possible for DB2 to work with concurrent I/O on
Linux? In some DB2 papers, I've read about DB2 "asynchoronous
I/O" on Linux; is that equivalent to concurrent IO on AIX?

-- Regards, Troels Arvin <tr****@arvin.dk>
http://troels.arvin.dk/

--
Haider