Exclusive Unix file system file write locks and Oracle and Veritas

Exclusive Unix file system file write locks and Oracle and Veritas

Post by NetComra » Fri, 10 Sep 1999 04:00:00



Hi,

I've read that having smaller file sizes on OLTP systems is good for
performance b/c of locks. But what about the Veritas File system? Is
there are some system variables I can change?

Thanx
---------------
Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
AOL: NetComrade  ICQ: 11340726 remove NSPAM to email

 
 
 

Exclusive Unix file system file write locks and Oracle and Veritas

Post by Paul Roger » Sat, 11 Sep 1999 04:00:00


Not really enough information to comment intelligently.  If I assume
some
things have been lost in translation then, using Veritas file system
with
a small stripe width like 16 or 32K can have a positive impact on Oracle

performance in an OLTP situation.  However, I want the stripe to cover
multiple (5 or more) disks to get the proper round-robin, no hot
spotting
effect.  I am unsure of the nature of the lock contention question.  It
turns
out that, depending upon the load on the system, Veritas can suffer from

mutex contention and as always you can get Oracle locks tied up in
knots.
I do not know if using Veritas files system will help the Oracle lock
contention (read I do not think that using Veritas will help Oracle lock

contention.)  Make sure that none of your OLTP transactions do full
table scans though.  That will severely limit the performance gains that

can be had with small stripe widths.

Paul Rogers
Sun Professional Services


> Hi,

> I've read that having smaller file sizes on OLTP systems is good for
> performance b/c of locks. But what about the Veritas File system? Is
> there are some system variables I can change?

> Thanx
> ---------------
> Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
> AOL: NetComrade  ICQ: 11340726 remove NSPAM to email


 
 
 

Exclusive Unix file system file write locks and Oracle and Veritas

Post by NetComra » Sat, 11 Sep 1999 04:00:00


What I really meant, what are the UNIX constraints on Oracle (or
should I say veritas constraints), since theoretically on a UFS there
are locks placed on a file upon writes. DOes Veritas do the same
thing? As for Oracle locking, I really don't have any on any of my
machines, expect for some MR locks, which are constantly there, and I
don't really know what they mean, but they are placed by PMON, so it
must be cleaning up something.



>Not really enough information to comment intelligently.  If I assume
>some
>things have been lost in translation then, using Veritas file system
>with
>a small stripe width like 16 or 32K can have a positive impact on Oracle

>performance in an OLTP situation.  However, I want the stripe to cover
>multiple (5 or more) disks to get the proper round-robin, no hot
>spotting
>effect.  I am unsure of the nature of the lock contention question.  It
>turns
>out that, depending upon the load on the system, Veritas can suffer from

>mutex contention and as always you can get Oracle locks tied up in
>knots.
>I do not know if using Veritas files system will help the Oracle lock
>contention (read I do not think that using Veritas will help Oracle lock

>contention.)  Make sure that none of your OLTP transactions do full
>table scans though.  That will severely limit the performance gains that

>can be had with small stripe widths.

>Paul Rogers
>Sun Professional Services


>> Hi,

>> I've read that having smaller file sizes on OLTP systems is good for
>> performance b/c of locks. But what about the Veritas File system? Is
>> there are some system variables I can change?

>> Thanx
>> ---------------
>> Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
>> AOL: NetComrade  ICQ: 11340726 remove NSPAM to email

---------------
Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
AOL: NetComrade  ICQ: 11340726 remove NSPAM to email
 
 
 

Exclusive Unix file system file write locks and Oracle and Veritas

Post by Jonathan Lewi » Sat, 11 Sep 1999 04:00:00


Under Unix file system, writers lock readers and readers lock
writers independently of anything that Oracle is doing.  The
lock is on the inode to ensure that a read cannot read data
that is changing.

This also happens under Veritas, but the nature of vxfs makes
the lock (usually) a little shorter.  Also the latest version of
veritas has further optimisation in it for 'quick db' which I think
helps further.

The main benefit for 'small files' is in mixing parallel query
and TEMP.  If you run create index in parallel, then all the
query slaves are likely to be reading and writing to TEMP
at the same time and interfering at the O/S level, so it
is better to have N files in temp matching the degree of
parallelism to reduce the probability of O/S collisions.

(This is the extreme case of course, but the same idea
holds any tablespace which is likely to be subject to
large numbers of concurrent reads and writes - Rollback
segment tablespaces are __likely__ to be the next
candidate on busy systems).

--

Jonathan Lewis
Yet another Oracle-related web site:  http://www.jlcomp.demon.co.uk


>What I really meant, what are the UNIX constraints on Oracle (or
>should I say veritas constraints), since theoretically on a UFS there
>are locks placed on a file upon writes. DOes Veritas do the same
>thing? As for Oracle locking, I really don't have any on any of my
>machines, expect for some MR locks, which are constantly there, and I
>don't really know what they mean, but they are placed by PMON, so it
>must be cleaning up something.

 
 
 

Exclusive Unix file system file write locks and Oracle and Veritas

Post by Thir » Sat, 11 Sep 1999 04:00:00


Could you Pls Clarify the statement ' writers lock Readers and readers
lock writers on the inode level in UFS ' ?

UFS uses File Buffer Cache which provides Read Consistent Data(similar to
ORacle) and they dont lock the entire Inode..

-Thiru


> Under Unix file system, writers lock readers and readers lock
> writers independently of anything that Oracle is doing.  The
> lock is on the inode to ensure that a read cannot read data
> that is changing.

> This also happens under Veritas, but the nature of vxfs makes
> the lock (usually) a little shorter.  Also the latest version of
> veritas has further optimisation in it for 'quick db' which I think
> helps further.

> The main benefit for 'small files' is in mixing parallel query
> and TEMP.  If you run create index in parallel, then all the
> query slaves are likely to be reading and writing to TEMP
> at the same time and interfering at the O/S level, so it
> is better to have N files in temp matching the degree of
> parallelism to reduce the probability of O/S collisions.

> (This is the extreme case of course, but the same idea
> holds any tablespace which is likely to be subject to
> large numbers of concurrent reads and writes - Rollback
> segment tablespaces are __likely__ to be the next
> candidate on busy systems).

> --

> Jonathan Lewis
> Yet another Oracle-related web site:  http://www.jlcomp.demon.co.uk


> >What I really meant, what are the UNIX constraints on Oracle (or
> >should I say veritas constraints), since theoretically on a UFS there
> >are locks placed on a file upon writes. DOes Veritas do the same
> >thing? As for Oracle locking, I really don't have any on any of my
> >machines, expect for some MR locks, which are constantly there, and I
> >don't really know what they mean, but they are placed by PMON, so it
> >must be cleaning up something.

 
 
 

Exclusive Unix file system file write locks and Oracle and Veritas

Post by Jonathan Lewi » Sat, 11 Sep 1999 04:00:00


Oracle uses OD_SYNC to do writes - which means
the write goes through the cache to the file.

When the write takes place there is an exclusive
lock placed on the inode entry for the duration of
the write.

--

Jonathan Lewis
Yet another Oracle-related web site:  http://www.jlcomp.demon.co.uk


>Could you Pls Clarify the statement ' writers lock Readers and readers
>lock writers on the inode level in UFS ' ?

>UFS uses File Buffer Cache which provides Read Consistent Data(similar to
>ORacle) and they dont lock the entire Inode..

>-Thiru


>> Under Unix file system, writers lock readers and readers lock
>> writers independently of anything that Oracle is doing.  The
>> lock is on the inode to ensure that a read cannot read data
>> that is changing.

 
 
 

1. Shrinking Veritas QIO files for Oracle backups or migrating QIO files to reg files

Hi, we are using Veritas HA for Oracle and in order to permantently
backup our databases I would like to shrink QIO files before backing
them up. Unfortunately Veritas wasn't able to provide me with an
adequate answer, so I am looking for an advice how to do it.

Migrating QIO files to regular files would probably be the simplest
way, but I don't know whether it is doable or not.

Please let me know if you have experienced a similar problem.

Thank you.

2. OpenLinux Stansard

3. Matrox Mystique ands X.

4. allowing dial in's etc...

5. Veritas/Oracle/Solaris 7 File System Specs.

6. RPC server initialization

7. Need help writing C program using UNIX system calls (read, write, etc) that copies files

8. how to install a AIX package

9. Enhanced Filing system - file system like DEC Advanced File system for Linux

10. sending email from oracle pl/sql or unix with out writing to a file

11. FIFO and exclusive file-locking over NFS ...?

12. temporary unique file with exclusive lock

13. File locking and data written to the end of a file