Re: Raw disk question

From: Gunther Birznieks <birznie_at_aplcenmp.apl.jhu.edu>
Date: Sat, 29 Oct 1994 00:05:33 GMT
Message-ID: <CyEqx9.7M4_at_aplcenmp.apl.jhu.edu>


tomhall_at_texaco.com (Tom Hall) writes:

>In article <37tupb$a5r_at_athos.cc.bellcore.com>, mjk_at_spanky.cc.bellcore.com (25994-kelly) says:
>>
>>Is the performance improvement from using raw disks only on writes or
>>does it improve tablespace scan type read operations as well?`
>>
>Using raw disk increases the performance for any type of disk access. Using raw devices elimates
>the need for the OS to interface with hardware interface level (i.e. in this case a file management
>system).

I would tend to think that READS would be increased in performance in a raw partition because of lack of the extra file system buffer overhead with the use of an equivalent size buffer for the database since the database can intelligently buffer data properly.

However, I would expect WRITES to tend to be slower because a file system does not know which writes to perform before others and can therefore come up with a faster way of optimizing all the writes that were submitted. However, a database program with raw partition access will schedule the writes in such a way that certain things get written first (like transaction log versus data pages that have been check pointed etc)...

Because of this, the Raw Partition has a greater reliability.. But the file system will be faster since it does not have to take into account the physical order the writes have to occur in to get a reliable system. Received on Sat Oct 29 1994 - 01:05:33 CET

Original text of this message