AIX Asynch. IO & DB2

AIX Asynch. IO & DB2

Post by cyberhabi.. » Thu, 03 Feb 2000 04:00:00



Pls forgive the repost, i'm cross-posting in comp.unix.aix in hopes i
can scare up an opinion... <G>


> Hi

> A DB2 instructor informed us that AIX Asynchronous IO (AIO) with UDB
is
> _NOT_ recommended, apparently because of the issue of _absolute_
> transaction recoverability, even w/ DMS raw logical volumes.

> The instructor seemed sure of his facts, but didn't provide any
> supporting documentation before he left (darn!). Can anyone comment on
> his claim?

> I understand DB2 does a lot of its own AIO in its IO servers, and I'm
> wondering if that's the crux of the issue.

> I _DID_ see a claim in this n.g. that asynch. IO is "a good thing"
> referencing a supposed redbook "Database Performance on AIX in DB2 UDB
> and Oracle Environments "SG24-551-00". Problem is I can't find that

> comment?)

> Using RLVs alone can increase performance by about a fifth, I'm
> wondering what kind of performance margin increase is to be gained
from
> AIO. Sybase requires it to install (as does, apparently, Oracle), but
> DB2 UDB doesn't require it.

> TIA,
> /leebert

Sent via Deja.com http://www.deja.com/
Before you buy.
 
 
 

AIX Asynch. IO & DB2

Post by cyberhabi.. » Thu, 03 Feb 2000 04:00:00


Pls forgive the repost, i'm cross-posting in comp.unix.aix in hopes i
can scare up an opinion... <G>


> Hi

> A DB2 instructor informed us that AIX Asynchronous IO (AIO) with UDB
is
> _NOT_ recommended, apparently because of the issue of _absolute_
> transaction recoverability, even w/ DMS raw logical volumes.

> The instructor seemed sure of his facts, but didn't provide any
> supporting documentation before he left (darn!). Can anyone comment on
> his claim?

> I understand DB2 does a lot of its own AIO in its IO servers, and I'm
> wondering if that's the crux of the issue.

> I _DID_ see a claim in this n.g. that asynch. IO is "a good thing"
> referencing a supposed redbook "Database Performance on AIX in DB2 UDB
> and Oracle Environments "SG24-551-00". Problem is I can't find that

> comment?)

> Using RLVs alone can increase performance by about a fifth, I'm
> wondering what kind of performance margin increase is to be gained
from
> AIO. Sybase requires it to install (as does, apparently, Oracle), but
> DB2 UDB doesn't require it.

> TIA,
> /leebert

Sent via Deja.com http://www.deja.com/
Before you buy.

 
 
 

AIX Asynch. IO & DB2

Post by Don Baug » Fri, 04 Feb 2000 04:00:00


I just went to www.redbooks.ibm.com and there was your redbook under the
just published section.  It was not hard at all to find.
---------
Don Baugh
Leveraged Solutions, Inc.


> Pls forgive the repost, i'm cross-posting in comp.unix.aix in hopes i
> can scare up an opinion... <G>


> > Hi

> > A DB2 instructor informed us that AIX Asynchronous IO (AIO) with UDB
> is
> > _NOT_ recommended, apparently because of the issue of _absolute_
> > transaction recoverability, even w/ DMS raw logical volumes.

> > The instructor seemed sure of his facts, but didn't provide any
> > supporting documentation before he left (darn!). Can anyone comment on
> > his claim?

> > I understand DB2 does a lot of its own AIO in its IO servers, and I'm
> > wondering if that's the crux of the issue.

> > I _DID_ see a claim in this n.g. that asynch. IO is "a good thing"
> > referencing a supposed redbook "Database Performance on AIX in DB2 UDB
> > and Oracle Environments "SG24-551-00". Problem is I can't find that

> > comment?)

> > Using RLVs alone can increase performance by about a fifth, I'm
> > wondering what kind of performance margin increase is to be gained
> from
> > AIO. Sybase requires it to install (as does, apparently, Oracle), but
> > DB2 UDB doesn't require it.

> > TIA,
> > /leebert

> Sent via Deja.com http://www.deja.com/
> Before you buy.

 
 
 

AIX Asynch. IO & DB2

Post by cyberhabi.. » Sat, 05 Feb 2000 04:00:00




Quote:> I just went to www.redbooks.ibm.com and there was your
>redbook under the
> just published section.  It was not hard at all to find.

You know? I have the same problem with my keys or finding the margarine
in the refrigerator... that'll teach me to use the search engine at
redbooks when looking would work just as well... <G>

thanx!

/leebert

Sent via Deja.com http://www.deja.com/
Before you buy.

 
 
 

1. VMM Contention/Scalability with IO on AIX/DB2?

First sorry for cross posting... but I thought it was likely that both
news groups could have relevant experience and comments.

 I am seeking some opinions about how well AIX scales IO.

Background: Hardware p690 16-way 48 GB mem
Applcation: DB2 EEE(partitioned 32 bit)

With our current database installation we have 8 logical nodes(db2) on
the same LPAR. We are using SMS filesystems for database storage. I
know we should be using RAW DMS tablespaces but the customer prefers
SMS. My question is since we are not using RAW IO is it probable that
the VMM for AIX will ever become a bottleneck for IO throughput as we
continue to add IO capacity (through adding controllers and more
drives)?.. We currently have about 250 MB/sec of random read IO
capacity and generally see about 30%-40% kernel time when pushing this
amount of IO through DB2. Currently the files are not MMAPed (So that
we can utilized all the memory because we are not using the 64 bit
version of db2 yet and DB2 immediately invalidates all pages in the
file cache loaded when using mmaped files). Obviously since we are
using non mmaped SMS tablespaces this kernel time is attributed to
double buffering and virtual memory mananagement(anything else?). Does
anyone have a feel for what percentage of kernel time would be spent
doing the double buffering and what percentage is spent doing the
virtual memory management. Will it be possible to overwhelm the vmm so
that it becomes a bottleneck. Is the vmm linearly scalable? That is,
if it is true that the VMM becomes a bottleneck will adding more CPU
solve the problem or is there some fundamental contention that makes
the scalability of the VMM with IO plateau out at some point?

Any insight is greatly appreciated..

Thanks for putting up with my rambling as I attempt to struggle with
putting the concepts in my question in words.

Thanks,
Scott

2. IPTABLES and Routing

3. Zoltrix Plus IO != 0x20c && IO != 0x30c

4. Questions of supported h/w

5. Asynch IO and Oracle

6. ide0=serialize

7. Asynch socket IO

8. Why is varaible always reset to zero?

9. Raw Devices and Asynch IO on Sybase/Linux

10. Q: Porting of asynch IO from SunOS

11. Asynch > TCP/IP > Asynch - How to?

12. F30 & 128-port Asynch Adapter

13. XML DTDs for synch & asynch interaction