cp_timer question

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

franky.leeuwerck@gmail.com - 25 Jan 2006 08:57 GMT

Hi,

I've been reading another thread in this group on extending the
transaction log and saw the discussion on the cp_timer parameter.

On Adv.Ingres 2.6 - hp-ux Itanium, we've been setting the
archiver_interval to 2 and cp_interval to 1. Transaction log size : 5
GB.

Two questions :
1) Is the cp_timer, just like the cp_interval, another trigger for
taking consistency points ?
2) Why does the cp_timer parameter occur in the recovery server section
and not in the section of the logging system ?

Regards,
Franky

Reply to this Message

Peter Gale - 25 Jan 2006 09:19 GMT

<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
Hi Fanky 
 
<blockquote
cite="mid1138179470.791865.316330@f14g2000cwb.googlegroups.com"
type="cite">
<pre wrap="">Hi,

I've been reading another thread in this group on extending the
transaction log and saw the discussion on the cp_timer parameter.

On Adv.Ingres 2.6 - hp-ux Itanium, we've been setting the
archiver_interval to 2 and cp_interval to 1. Transaction log size : 5
GB.

Two questions :
1) Is the cp_timer, just like the cp_interval, another trigger for
taking consistency points ?
</pre>
</blockquote>
Yes, which ever one is reached first triggers the CP. 
<blockquote
cite="mid1138179470.791865.316330@f14g2000cwb.googlegroups.com"
type="cite">
<pre wrap="">2) Why does the cp_timer parameter occur in the recovery server section
and not in the section of the logging system ?
</pre>
</blockquote>
cp_timer is triggered by a thread in the recovery server, hence the
parameter is only needed there. If set to 0 the thread does not run. 
<blockquote
cite="mid1138179470.791865.316330@f14g2000cwb.googlegroups.com"
type="cite">
<pre wrap="">

Regards,
Franky

_______________________________________________
Info-ingres mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Info-ingres@cariboulake.com">Info-ingres@cariboulake.com</a>
<a class="moz-txt-link-freetext" href="http://mailman.cariboulake.com/mailman/listinfo.py/info-ingres">http://mailman.c
ariboulake.com/mailman/listinfo.py/info-ingres</a>

</pre>
</blockquote>
<pre>-- </pre>
Peter Gale 
Comprehensive Solutions (US) 
<o:p></o:p>Comprehensive Solutions International (UK) 
T: +44
(0)1398
341777               
M: +44 (0)7831 513181 
<a
href="mailto:PGale@Comp-Soln.co.uk">PGale@Comp-Soln.co.uk</a>  
<a href="http://www.Comp-Soln.co.uk"
title="http://www.comp-soln.co.uk/">www.Comp-Soln.co.uk</a><o:p></o:p>
</body>
</html>

Reply to this Message

franky.leeuwerck@gmail.com - 25 Jan 2006 09:28 GMT

Hi Peter,

Thanks for the input.

Would it make sense to have both parameters available in the logging
and in the recovery section of the configuration file ?
I guess so.

But this is not the case now. I can only guess, that both were
implemented for different purposes and on different timings in life of
the product.

Franky

Peter Gale schreef:

> Hi Fanky
>
[quoted text clipped - 32 lines]
> T: +44 (0)1398 341777 M: +44 (0)7831 513181
> PGale@Comp-Soln.co.uk www.Comp-Soln.co.uk

Reply to this Message

Peter Gale - 25 Jan 2006 10:03 GMT

<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
Hi Franky (sorry for the earlier typo) 
 
I would agree that it would make sense to have the 2 parameters in the
same part of the configuration but clearly it was an architectural
thing that dictated this. 
 
On a related point I would advocate the use of cp_timer as well as
cp_interval (or cp_interval_mb on R3). The frequency of consistency
points is fundamental in determining the guaranteed recovery point for
your databases. In other words if you have a business requirement to
guarantee that your databases can be recovered to within N minutes of
the failure time for ANY failure, then you need to ensure that CP's and
the subsequent Archive Cycle are going to occur (and finish) at least
every N minutes and probably more frequently. If cp_timer is not set
then neither cp_interval nor cp_interval_mb can give you that guarantee. 
 
Peter 
 
<blockquote
cite="mid1138181333.490572.233770@g43g2000cwa.googlegroups.com"
type="cite">
<pre wrap="">Hi Peter,

Thanks for the input.

Would it make sense to have both parameters available in the logging
and in the recovery section of the configuration file ?
I guess so.

But this is not the case now. I can only guess, that both were
implemented for different purposes and on different timings in life of
the product.

Franky

Peter Gale schreef:

</pre>
<blockquote type="cite">
<pre wrap="">Hi Fanky

Hi,

I've been reading another thread in this group on extending the
transaction log and saw the discussion on the cp_timer parameter.

On Adv.Ingres 2.6 - hp-ux Itanium, we've been setting the
archiver_interval to 2 and cp_interval to 1. Transaction log size : 5
GB.

Two questions :
1) Is the cp_timer, just like the cp_interval, another trigger for
taking consistency points ?

Yes, which ever one is reached first triggers the CP.

2) Why does the cp_timer parameter occur in the recovery server section
and not in the section of the logging system ?

cp_timer is triggered by a thread in the recovery server, hence the parameter is only needed there. If set to 0 the thread does not run.

Regards,
Franky

_______________________________________________
Info-ingres mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Info-ingres@cariboulake.com">Info-ingres@cariboulake.com</a>
<a class="moz-txt-link-freetext" href="http://mailman.cariboulake.com/mailman/listinfo.py/info-ingres">http://mailman.c
ariboulake.com/mailman/listinfo.py/info-ingres</a>

--
*Peter Gale*
Comprehensive Solutions (US)
Comprehensive Solutions International (UK)
T: +44 (0)1398 341777 M: +44 (0)7831 513181
<a class="moz-txt-link-abbreviated" href="mailto:PGale@Comp-Soln.co.uk">PGale@Comp-Soln.co.uk</a> <a class="moz-txt-link-abbreviated" href="http://www.Comp-Soln.co.uk">www.Comp-Soln.co.uk</a>
</pre>
</blockquote>
<pre wrap="">
_______________________________________________
Info-ingres mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Info-ingres@cariboulake.com">Info-ingres@cariboulake.com</a>
<a class="moz-txt-link-freetext" href="http://mailman.cariboulake.com/mailman/listinfo.py/info-ingres">http://mailman.c
ariboulake.com/mailman/listinfo.py/info-ingres</a>

</pre>
</blockquote>
</body>
</html>

Reply to this Message

martin.bowes@ctsu.ox.ac.uk - 25 Jan 2006 09:29 GMT

Hi franky,

> On Adv.Ingres 2.6 - hp-ux Itanium, we've been setting the
> archiver_interval to 2 and cp_interval to 1. Transaction log size : 5
[quoted text clipped - 3 lines]
> 1) Is the cp_timer, just like the cp_interval, another trigger for
> taking consistency points ?

Yes.

2) Why does the cp_timer parameter occur

> in the recovery server section and not in the section of the logging
> system ?

Good question - damned if I know.

Martin Bowes
--
Random Duckman Quote #71:
Ajax: Pardon me, Tammy and I need some canola oil, some rope, and
a zucchini.

Reply to this Message

franky.leeuwerck@gmail.com - 26 Jan 2006 08:46 GMT

Ok Folks,

Thanks for all your inputs.
The cp_timer parm looks interesting.
Next thing that I want to do is to start using it.
Can there be a reason why I shouldn't, like performance degradation for
the logging system ?

Franky

Reply to this Message

Roy Hann - 26 Jan 2006 09:09 GMT

> Ok Folks,
>
[quoted text clipped - 3 lines]
> Can there be a reason why I shouldn't, like performance degradation for
> the logging system ?

Well you probably don't want to go mad with it. Apart from being moderately
intensive (though not as intensive as it used to be), a consistency point
posts all modified pages to disk, regardless of whether the update has been
committed or whether all the updates to a page are finished. Doing
excessively frequent CPs would mean having more rolled-back updates to undo
on disk, and doing multiple posts to disk to capture all the updates to a
page. On a very busy system you might notice that. (It would have to be
very, very busy though.)

You could spend a whole lot of time testing and measuring and re-testing, to
get optimal performance. But on most systems I'd say to heck with it, I
just want to make sure my undo/redo recovery time after a crash is
guaranteed to be less than five minutes and set cp_timer to 300.

Roy Hann (rhann at rationalcommerce dot com)
Rational Commerce Ltd.
www.rationalcommerce.com
"Ingres development, tuning, and training experts"
Ingres Corporation Partner

Reply to this Message

franky.leeuwerck@gmail.com - 27 Jan 2006 09:01 GMT

Roy,

Thanks for your inputs !
I found this a very useful discussion thread.

Franky

Reply to this Message