RE: segment growth
- From: "Mark W. Farnham" <mwf@xxxxxxxx>
- To: <remigiusz.sokolowski@xxxxxxxxxx>, "'oracle-l'" <oracle-l@xxxxxxxxxxxxx>
- Date: Wed, 2 Apr 2014 08:11:06 -0400
As with any trend analysis you can do something like least squares to make a
projection.
If you want to have an idea whether the trend produced is somewhat reliable,
you then have to do a confidence calculation for the type of trend calculation
you made.
Few points lead to low confidence, few points combined with data that are bits
of a saw tooth wave pattern lead to oscillating projections. Consider a series
1,3,1,3 on the amplitude axis against uniformly increasing time axis (say once
a day for storage) repeating and calculate least square with even and odd
number of points included. Graphs ending in 3 will project growth, those ending
in 1 will project. With enough points the trend will approach zero slope at an
amplitude of very near 2, but it would still oscillate slightly.
So you have to be very careful about values that rise and fall.
You also have to be very careful about anticipation of epochal events that make
the history of points irrelevant.
For capital planning, you probably should ignore shrinkage unless it is
permanent and over some threshold.
To do this, the simplest think is to keep two values: How big am I today and
max value. (Update max value if current is bigger than max).
For load/truncate tables (or anything else that has a pattern like that), you
need to check the max size just before truncate to avoid surprises.
And then perhaps see how much overestimate you have from the drift between
current and max over purchase planning forecast windows like a quarter or a
year.
Oh – an don’t look at the data until you’ve got a reasonable number of points.
It will just annoy you.
mwf
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On
Behalf Of Remigiusz Sokolowski
Sent: Wednesday, April 02, 2014 2:18 AM
To: oracle-l
Subject: segment growth
hi,
I have just tried to realize how to count a segment growth trend based on the
dba_hits_seg_stat rows.
And now I am puzzled:
1. on the metalink I have found the article 1395195.1, which suggests to simply
sum the db_block_changes_delta
2. I displayed the data for some particular segment to find out the possible
relations between numbers and I would say this is rather the value, which
indicates literally the delta of block changes i.e. all the deletes, inserts,
updates and not only new blocks - for example because the
db_block_changes_delta is always positive and db_block_changes_total always
grows
3. So I look at space_used_total assuming this is the total space used by the
segment in the moment a snap is taken - and here longer example
current size of the object
-----------------
object_name | blocks | bytes
ZABBIX.ITEMS 4992 40894464
excerpt from snap values - few subsequent snaps
-----------------
OBJ_SNAP, SPACE_USED_TOTAL, SPACE_USED_DELTA,
13356 933731 9503
13357 937827 4096
13358 940940 3113
13359 -3277 -3277
13360 737 4014
13361 6062 5325
13362 7045 983
until 13358 all is fine I see the logic (
space_used_total(n)=space_used_total(n-1)+space_used_delta(n) )
but then negative total value (and the logic's gone) - no idea what to think
about that.
And the main question - how to calculate the segment growth trend in a reliable
manner?
Best regards
Remigiusz
--
Pole na kazi
----------------------------------------------------------------------
Remigiusz Sokolowski <mailto:remigiusz.sokolowski@xxxxxxxxxx>
<remigiusz.sokolowski@xxxxxxxxxx>
pos : Senior DBA at DIiUSI
comp : Nordea IT Polska Sp. z o.o.
addr : Luzycka 6A st, 81-537 Gdynia, Poland
phone : +48 58 667 17 43
mobile: +48 602 42 42 77
geo : 54.4952N 18.4332E
Nordea IT Polska Sp.z o.o. z siedzibą w 81-537 Gdynia, ul. Łużycka 6A wpisaną
do Rejestru Przedsiębiorców Krajowego Rejestru Sądowego pod numerem:
0000429783, dla której dokumentację przechowuje Sąd Rejonowy dla M.St. Warszawy
w Warszawie, XII Wydział Gospodarczy Krajowego Rejestru Sądowego, o kapitale
zakładowym i opłaconym w wysokości: 40.000.000,00 złotych, NIP: 527-268-29-50,
REGON: 146278110
-- //www.freelists.org/webpage/oracle-l
Other related posts:
