Re: Re: HCC in Pillar Storage

From: Tanel Poder <tanel_at_tanelpoder.com>
Date: Tue, 20 Mar 2012 19:56:15 +0000
Message-ID: <CAMHX9J+KsHOFPS4rnZn9p7eBPdWrSftH7kwx4FSzUoKM1VDiMw_at_mail.gmail.com>

I wouldn't do that :)
But if any of the datafiles in a tablespace are outside HCC-supported storage, then the entire tablespace will become not supported for HCC (nor smart scanning).

So watch out before adding NFS-based datafiles into your existing storage cell based tablespaces, you will suddenly start getting these errors when trying to access any HCC compressed data:

*ORA-64307: hybrid columnar compression is not supported for tablespaces on this storage type*

... and the smart scans are disabled for the entire tablespace.

Starting from 11g, you can actually drop individual datafiles of a tablespace (if they are empty), so once you drop the non-compatible file, everything will be back to normal.

--
*Tanel Poder*
Enkitec Europe
+372 56 956 181
http://www.enkitec.com/
Expert Oracle Exadata book:
http://www.apress.com/9781430233923


On Tue, Mar 20, 2012 at 5:46 PM, Ls Cheng <exriscer_at_gmail.com> wrote:



> Thanks for the explanation

>

> So if a tablespace has 2 datafiles, one per array, and if a segment

> located in the tablespace spans in these two datafiles, can this segment

> have part of data in HCC and others parts not?

>

> Thanks!

>

>

>

> On Tue, Mar 20, 2012 at 6:29 PM, Tanel Poder <tanel_at_tanelpoder.com> wrote:

>

>> Oracle supports HCC on Exadata and from 11.2.0.3 onwards on Oracle's (Sum)

>> ZFS appliance and Pillar. Obviously on non-Exadata the *de*compression

>>

>> always happens at the database layer (on Exadata it depends on a number of

>> things). The compression always happens in the database layer on both

>> Exadata and non-Exadata.

>> The HCC "support" is decided at the tablespace level, but HCC itself is

>> enabled at the segment level. So you can have a table residing in

>> HCC-capable tablespace, but you decide not to compress it.

>>

>> This segment-level approach also means that you can move the oldest

>> partitions of a partitioned fact table onto Pillar/ZFS appliance (and keep

>> it compressed with HCC if you want) while keeping the newest partitions of

>> the same table on Exadata storage. By the way, even with regular NFS

>> appliances or iSCSI storage you can keep some partitions of a table off

>> Exadata (without HCC) while keeping other partitions on Exadata.

>>

>> --

>> *Tanel Poder*

>>

>> Enkitec Europe

>> +372 56 956 181

>> http://www.enkitec.com/

>> Expert Oracle Exadata book:

>> http://www.apress.com/9781430233923

>>

>>

>>

>> On Tue, Mar 20, 2012 at 4:13 PM, Vit Spinka <vit.spinka_at_vitspinka.cz>

>> wrote:

>>

>> >

>> > The HCC is usually done by the storage; if the storage decides not to do

>> > so (e.g. due to CPU load), it will pass the blocks uncompressed and it's

>> > then job of the database to uncompress them.  So Oracle does not support

>> > HCC on non-Exadata/Pillar, but it could do so, but with performance hit.

>> >

>> > >>  I am wondering because I am not very sure where is HCC done, if it

>> is

>> > done

>> > >>  in the DB how the hcc algorithm applies for some blocks (pillar) and

>> > some

>> > >>  not (EVA), if it works shouldnt be a performance hit?

>> > >>

>> > >

>> >

>> > --

>> > http://www.freelists.org/webpage/oracle-l

>> >

>> >

>> >

>>

>>

>> --

>> http://www.freelists.org/webpage/oracle-l

>>

>>

>>

>



--
http://www.freelists.org/webpage/oracle-l

Received on Tue Mar 20 2012 - 14:56:15 CDT