Re: When does Oracle use 'Index Fast Scan'

From: David Hau <davehau123_at_netscape.net>
Date: Tue, 27 Jan 2004 08:14:27 -0800
Message-ID: <F001.005DE349.20040127081427@fatcity.com>

This is where the access time of your disks (or SAN) makes a difference.   If your disks have really fast access time, then a random-access pattern would not cause much performance degradation and so a range scan would not be slow at all, even though it's traversing the b-tree index structure. If you're only striping together disks with relatively slow access time (e.g. using a striped IDE disk array), then you have high throughput but not that fast an access time. In this case, fast full index scan would be much faster than an index range scan because the fast full scan reads the blocks sequentially and a sequential disk I/O requires only positioning the head once (assuming the disk is not fragmented). The rest of the time depends on the throughput. If you stripe together a large enough number of IDE disks, then your throughput is great but your access time is still the access time of a single IDE drive which is not that fast.

This is assuming you need to do a physical I/O to obtain the blocks. Of course, if the blocks already reside in the buffer cache, then it's a different story.

Regards,
Dave

ryan.gaffuri_at_cox.net wrote:

> btw, in many cases range scan is faster than a fast full scan. Range scan recursively hits the nodes that are needed and skips the ones that are not. So it reads less blocks. 
> 
> So if you are looking for a 'range' or a specific value, range scan beats fast full scan most of the time. Less Logical and Physical I/Os. 
> 
> test it and hint your queries
> 
> 
> 

>>From: David Hau <davehau123_at_netscape.net>
>>Date: 2004/01/26 Mon PM 10:34:25 EST
>>To: Multiple recipients of list ORACLE-L <ORACLE-L_at_fatcity.com>
>>Subject: Re: When does Oracle use 'Index Fast Scan'
>>
>>Correction: the Index Range Scan can be parallelized when it involves
>>multiple partitions.
>>
>>- Dave
>>
>>
>>David Hau wrote:
>>
>>
>>>I assume you're talking about the Fast Full Index Scan. This is used
>>>when the index contains all the columns necessary to answer the query.
>>>
>>>It's faster than a Full Table Scan because indexes are smaller than
>>>entire rows, so a Fast Full Index Scan will scan fewer blocks than a
>>>Full Table Scan.
>>>
>>>It's faster than an Index Range Scan firstly because Fast Full Index
>>>Scan scans the blocks in sequential order, whereas the Index Range
>>>Scan traverses the B-tree index structure in scanning the blocks,
>>>resulting in a random access I/O pattern which is slower. This is
>>>also why the Oracle documentation says that with a Fast Full Index
>>>Scan, the result is not sorted by the index key (because the result is
>>>not obtained by traversing the index structure.) Secondly, the better
>>>performance is also because the Fast Full Index Scan uses multiblock
>>>reads and is capable of parallel operation, whereas the Index Range
>>>Scan is capable of neither.
>>>
>>>Regards,
>>>Dave.
>>>
>>>
>>>
>>>ryan.gaffuri_at_cox.net wrote:
>>>
>>>
>>>>I have found that the vast majority of time that Oracle chooses this
>>>>method, my statistics are stale and the query is sub-optimal. One
>>>>time, Oracle changed from a 'range scan' to this type of scan with a
>>>>FIRST_ROWS hint and this reduced performance.
>>>>
>>>>This is just a full scan of the index, one block at a time right?
>>>>When would this ever be superior to a Fast Full Scan or a Range Scan?
>>>
>>>
>>>
>>--
>>Please see the official ORACLE-L FAQ: http://www.orafaq.net
>>--
>>Author: David Hau
>> INET: davehau123_at_netscape.net
>>
>>Fat City Network Services -- 858-538-5051 http://www.fatcity.com
>>San Diego, California -- Mailing list and web hosting services
>>---------------------------------------------------------------------
>>To REMOVE yourself from this mailing list, send an E-Mail message
>>to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
>>the message BODY, include a line containing: UNSUB ORACLE-L
>>(or the name of mailing list you want to be removed from). You may
>>also send the HELP command for other information (like subscribing).
>>

>
>

-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.net
-- 
Author: David Hau
  INET: davehau123_at_netscape.net

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).

Received on Tue Jan 27 2004 - 10:14:27 CST