Hi Dmitry,

Take your point on there being no orphaned records, default for
TempCacheLimit is 8 MB (see below dragged from conf file). And proof in
terms of screen shots are below. You may also see on there FB2.5 which has
many more websites on it.

fwiw, I am just trying to understand why the mem *appears* so high before I
port another 10 to 15 databases over to v3.

regards

Si

Straight from conf file:


#
# The maximum amount of the temporary space that can be cached
# in memory.
#
# For Classic servers, this setting is defaulted to 8 MB.
# Although it can be increased, the value applies to each client
# connection/server instance and thus consumes a lot of memory.
#
# Type: integer
#
#TempCacheLimit = 64M


[image: Inline images 3]
[image: Inline images 2]

On 7 November 2016 at 08:41, Dmitry Yemanov dimitr@...
[firebird-support] <firebird-support@yahoogroups.com> wrote:

>
>
> 04.11.2016 13:30, Si Carter wrote:
>
> > The memory usage crept up to just under 3gb again
>
> Once again, where do you get that number from? Task manager? Perfmon?
> top? Screenshot proof would be appreciated.
>
> > The memory usage crept up to just under 3gb again, and I took a snapshot
> > of the data using the previous sql, from all 4 databases. Between the 4
> > databases there are over 1500 orphaned records, and between the 4
> > databases there are 30 identifiable connections.
>
> There are no "orphaned" records, it's just a result of your outer join
> -- you also get non-identified transaction- and statement-level counters
> (mon$stat_group > 1).
>
> And because of that, you shouldn't sum all the values in the column -
> the counters are already aggregated
> statement->transaction->attachment->database, so you just get wrongly
> multiplied results.
>
> > I have placed the raw data in a csv file (http://tinyurl.com/zkqf9bo)
>
> The current memory usage is just 350MB but there was a peak of 3.4GB
> (3GB of them were used for db1.fdb).
>
> What is your TempCacheLimit setting in firebird.conf? If it's set too
> high, the peaks could be explained by heavy sorts.
>
> Dmitry
>
>
>

[Non-text portions of this message have been removed]