Den 2014-02-28 00:36 skrev Dmitry Kuzmenko såhär:

>
> Hello, Kjell!
>
> Friday, February 28, 2014, 3:23:34 AM, you wrote:
>
> KR> I'm going to do a gbak backup/restore on my 100 Gbyte database this
> KR> weekend, and was wondering if perhaps I should bump up the page size
> KR> from 8 kbyte to 16 kbyte.
>
> KR> What should I look for?
>
> KR> I'm running gstat -a -i right now. Is this the appropriate options to
> KR> get the relevant figures?
>
> No. The best way, always, minimum gsta -h, or maximum gstat -r.
> Only gstat -r shows record size, versions, etc.
> You should look at indices, that have depth greater than 3 (equal to 4
> or more).
>

OK, I ran gstat -r and found a few indices with depth = 4, as follows:

-----------------------------------------------
Företag (146)
Primary pointer page: 231, Index root page: 232
Average record length: 292.38, total records: 2462168
Average version length: 0.00, total versions: 0, max versions: 0
Data pages: 101698, data page slots: 101698, average fill: 92%
Fill distribution:
0 - 19% = 0
20 - 39% = 0
40 - 59% = 1
60 - 79% = 3
80 - 99% = 101694

-- Not searched very often. Inserts are not time critical. Probably
not an issue.
Index IX_Företag_NamnCI (22)
Depth: 4, leaf buckets: 22405, nodes: 2462168
Average data length: 34.17, total dup: 345654, max dup: 968
Fill distribution:
0 - 19% = 58
20 - 39% = 15
40 - 59% = 20446
60 - 79% = 28
80 - 99% = 1858

-----------------------------------------------
Uppgift (172)
Primary pointer page: 284, Index root page: 285
Average record length: 77.90, total records: 208909973
Average version length: 0.00, total versions: 0, max versions: 0
Data pages: 3049518, data page slots: 3049518, average fill: 80%
Fill distribution:
0 - 19% = 0
20 - 39% = 0
40 - 59% = 0
60 - 79% = 2001636
80 - 99% = 1047882

-- Heavily used for both searches and inserts.
-- A performance issue here would have an impact. Most critical for
reading.
Index IX_PK_Uppgift (0)
Depth: 4, leaf buckets: 314034, nodes: 208909973
Average data length: 5.01, total dup: 0, max dup: 0
Fill distribution:
0 - 19% = 47
20 - 39% = 0
40 - 59% = 5952
60 - 79% = 2785
80 - 99% = 305250

-- Heavily used for both searches and inserts.
-- A performance issue here would have an impact. Most critical for
reading.
Index IX_Uppgift_Hållare (2)
Depth: 4, leaf buckets: 434584, nodes: 208909973
Average data length: 3.95, total dup: 45804696, max dup: 998003
Fill distribution:
0 - 19% = 555
20 - 39% = 0
40 - 59% = 316053
60 - 79% = 1476
80 - 99% = 116500
-----------------------------------------------
Uppgiftshållare (175)
Primary pointer page: 290, Index root page: 291
Average record length: 42.23, total records: 166793365
Average version length: 0.00, total versions: 0, max versions: 0
Data pages: 1689968, data page slots: 1689968, average fill: 72%
Fill distribution:
0 - 19% = 1
20 - 39% = 0
40 - 59% = 0
60 - 79% = 1689967
80 - 99% = 0

-- Heavily used for both searches and inserts.
-- A performance issue here would have an impact. Most critical for
reading.
Index IX_PK_Uppgiftshållare (0)
Depth: 4, leaf buckets: 247620, nodes: 166793365
Average data length: 5.05, total dup: 0, max dup: 0
Fill distribution:
0 - 19% = 1
20 - 39% = 0
40 - 59% = 141
60 - 79% = 0
80 - 99% = 247478

-- Heavily used for both searches and inserts.
-- A performance issue here would have an impact. Most critical for
reading.
Index IX_Uppgiftshållare_HanDS2 (1)
Depth: 4, leaf buckets: 155606, nodes: 166793365
Average data length: 0.48, total dup: 151362720, max dup: 46
Fill distribution:
0 - 19% = 144
20 - 39% = 1
40 - 59% = 40447
60 - 79% = 619
80 - 99% = 114395

-- Heavily used for both searches and inserts.
-- A performance issue here would have an impact. Most critical for
reading.
Index UppgiftshållareUnika (3)
Depth: 4, leaf buckets: 424641, nodes: 166793365
Average data length: 10.97, total dup: 0, max dup: 0
Fill distribution:
0 - 19% = 1669
20 - 39% = 0
40 - 59% = 99013
60 - 79% = 2327
80 - 99% = 321632

So, based on that gstat info and my comments for each, would you
recommend to try 16 kbyte page size?

> Anyway, you may always compare performance. 100gb is not such a big
> database that could not be set to 8k back at appropriate time if
> you will find that something become slower at 16k.
>

I know it's not to be regarded as very large compared to many others,
but it's large enough to require careful choices to get adequate
performance on our rather low budget server. :-)

Regards,
Kjell