>On Monday 28 June 2004 08:16, Santiago Cordoba wrote:
>
> >
> > Oldest active 5396133
> > Next transaction 5400951

At 09:49 AM 28/06/2004 +0000, Phil Shrimpton wrote:

>You have 4818 active transactions in your system. Which is going to cause a
>performance issue at best

Not necessarily 4818 active transactions...it means that 4817 transactions
have started since TID 5396133 started; and the OAT hasn't finished
yet. This means that garbage from those intervening transactions just
accumulates and doesn't get cleaned up. It's not the retained garbage, per
se, that's causing the system to buckle and break, but the buildup of row
references in memory.

Also this:

> Oldest active 5396133
>
> Oldest snapshot 5394502

indicates there was a stuck transaction when TID 5396133 started. Oldest
snapshot indicates the "high-water mark" of the last garbage collection. If
this is the way the system is allowed to roll, it suggests some attention
to hard committing is in order.

Do you have user tasks around that are using a Borland-style "Autocommit"
transaction (Post + CommitRetaining in a single hit for each DML operation,
typically operating on rows in a scrollable dataset) and never performing a
hard commit? That's what your figures could indicate...

/heLen