RE: db and storage snapshot - may be a little OFF-TOPIC
- From: Jonathan Lewis <jonathan@xxxxxxxxxxxxxxxxxx>
- To: "remigiusz.sokolowski@xxxxxxxxxx" <remigiusz.sokolowski@xxxxxxxxxx>, oracle-l <oracle-l@xxxxxxxxxxxxx>
- Date: Fri, 28 Jun 2013 09:08:10 +0000
The immediate response to this is to look at indexes.
Given the right (which means "unlucky") index, you could find that inserting
1,000 rows into a table updates 1,000 blocks in the index. This won't quite
match your description - but if you have a few very large indexes on a couple
of critical table it could easily result in the snapshot size becoming a large
fraction of the database size.
(There is an option in the latest versions of dbv that allows you to specify a
"high_scn" when you check the database, and the output shows you the number of
blocks changed since that SCN for each file - this might be a long report, but
might give you some interesting information. See:
http://iiotzov.wordpress.com/2013/05/14/measuring-the-benefits-of-database-virtualizationthin-provisioning-solutions/
Regards
Jonathan Lewis
________________________________________
From: oracle-l-bounce@xxxxxxxxxxxxx [oracle-l-bounce@xxxxxxxxxxxxx] on behalf
of Remigiusz Sokolowski [remigiusz.sokolowski@xxxxxxxxxx]
Sent: 28 June 2013 09:24
To: oracle-l
Subject: db and storage snapshot - may be a little OFF-TOPIC
hi,
we support test env creation partially by making snapshots on the
storage array level. One of the nice sides of this are storage savings.
The algorithm behind snapshot is simple - modification of the snapshot
copies modified block content to snapshot storage and modifies it there,
while source stays intact.
We found few cases where the storage designated for snapshot
modification become equal to the snapshot source, which means every
resource block modified - all of them Oracle databases.
Are You able to name a database operation, which would modify every
block in database (or assuming this is about array blocks (in this case
256k) one per every 32 blocks)?
Of course there are no operations like database restore and no visible
pattern despite the fact that only Oracle databases are prone to it (but
I would not say this is 100% sure). The array is IBM Storwize. The db is
11.2. The OS is Linux flavour.
I appreciate the slightest hypothesis, thanks in advance
Best regards
Remigiusz
--
Pole na kazi
----------------------------------------------------------------------
Remigiusz Sokolowski <remigiusz.sokolowski@xxxxxxxxxx>
pos : Senior DBA at DIiUSI
addr : Nordea Bank Polska SA, Luzycka 6A st, 81-537 Gdynia, Poland
phone : +48 58 667 17 43
mobile: +48 602 42 42 77
Nordea Bank Polska S.A. z siedziba w Gdyni, ul. Kielecka 2, 81-303 Gdynia,
wpisana do Rejestru Przedsiebiorców Krajowego Rejestru Sadowego pod numerem:
0000021828,
dla której dokumentacje przechowuje Sad Rejonowy Gdansk - Pólnoc w Gdansku,
VIII Wydzial Gospodarczy Krajowego Rejestru Sadowego,
o kapitale zakladowym i wplaconym w wysokosci: 277.493.500,00 zlotych,
NIP: 586-000-78-20, REGON: 190024711--
//www.freelists.org/webpage/oracle-l--
//www.freelists.org/webpage/oracle-l
Other related posts:
