Re: Hard numbers for INITRANS and MAXTRANS

On Tue, 01 Aug 2000 13:59:55 GMT, vasarpota_at_my-deja.com wrote:

>I understand that INITTRANS and MAXTRANS should be set object by
>object. Ia also understand that the number are determined by the size
>of the object and the concurrency for the object. But what hard values
>are reccomended?

Leave them defaulted. Works 99% of the time. they are not related to size of object but to concurrency.

>
>I am also assuming that indexes have a higher INITRANS and MAXTRANS
>than data segments.

Not necessarily. The default may be different, though.

>
>Do you usually put INITRANS=MAXTRNAS for the more static tables? Is it
>good practive to always keep the two equal?

No. And most definitely no.

MAXTRANS is like a high-water mark. Only to be used if ever the number of transactions updating rows in a given block goes near its value. I don't know of any case where that has actually happened using the default MAXTRANS. Usually the block header ends first...

INITRANS reserves "transaction slots" in the block header before they are really asked for. Like a sort of a "cache". One of those is used for every row locked in the block by a given transaction - and therefore for every row potentially updated. If you have many processes updating in the same block, you may want a higher number for INITRANS so the kernel doesn't have to get them dynamically.

I've seen this to be useful in cases where multiple load processes are adding to the end of a table. And some very odd OLTP cases. Other than that, never seen it needed.

Cheers
Nuno Souto
nsouto_at_nsw.bigpond.net.au.nospam
http://www.users.bigpond.net.au/the_Den/index.html Received on Wed Aug 02 2000 - 00:00:00 CDT