RE: Raw Devices v.s. File System

From: Gaja Krishna Vaidyanatha <gajav_at_yahoo.com>
Date: Wed, 12 Jul 2000 10:46:01 -0700 (PDT)
Message-Id: <10556.111841@fatcity.com>

Patrice,

I have done some benchmarks on "Raw vs. Unix ufs filesystems", but it was a while ago (5 years). At that time, the difference in performance was significant. Raw was clearly better and in some cases by a factor of 40% compared with "ufs filesystems". The primary reason for not having done any benchmarks lately, is the prevelant availability and use of advanced filesystems such as vxfs, xfs, jfs (as the case may be) on most flavors of Unix. Point is - why benchmark something that you don't use anymore? In my world, if I have a choice to use an advanced filesystem, I pick that over ufs.

But, the more important point that needs to be made here is that "filesystems" have come a long way from their "ufs" version. In the past 5 years, the 40% performance difference has significantly shrunk by the advent of the advanced filesystems which provide intelligent I/O operations, support for "direct i/o" on some platforms, better journalling and recoverability.

Given that the performance difference between these advanced fileystems and Raw is not that much, it really begs the issue as to why one would want to give up the benefits of a filesystem. Unless of course, one is running Oracle Parallel Server, where it is currently not an option. I did hear some "rumors" that OPS may be supported on filesystems in some future release. Don't know any more than that.

Yes, there are specific configuration details that you need to pay attention to - like configuring the filesystem blocksize to be equal to db_block_size, setting the stripewidth appropriately, understanding your hardware's I/O bandwidth limitations etc.

But the benchmarks that I have done recently (in the past 24 months) on production systems with significant amounts of data and transactions, tell me a great story. The comparision of "raw with asynch i/o" vs. "xfs with direct i/o" and "raw with asynch i/o" vs. "vxfs with Quick i/O", have me really convinced that advanced filesystems are here to stay and provide raw-comparable performance for most hybrid (OLTP + Batch) and DSS/Data Warehousing environments.

Cheers,

Gaja.

• "Boivin, Patrice J" <BoivinP_at_mar.dfo-mpo.gc.ca> wrote:
  > I read those articles, the theory sounds good.
  >
  > When I tested this on NT, however, I wanted to bypass the NT
  > filesystem to
  > try to speed it up. I figured the NT kernel is much larger
  > than the UNIX
  > one, so if there is a difference, it should be more noticeable
  > on the NT
  > machine.
  >
  > My tests show no difference whatsoever between raw and ntfs
  > volumes. This
  > could be because the performance difference is too small to
  > measure anyway,
  > or because I/O speed is limited by other factors (like a slow
  > controller,
  > the drives on my test server are only IDE drives) and so the
  > raw vs. ntfs
  > makes no difference.
  >
  > If someone did tests to compare UNIX raw partitions vs. UNIX
  > ufs partitions,
  > I would like to learn whether they found an actual difference.
  >
  > I know now that raw partitions on nt provide minimal to no
  > benefit on my NT
  > test server.
  >
  > Regards,
  > Patrice Boivin
  > Systems Analyst (Oracle Certified DBA)
  >
  > Systems Admin & Operations | Admin. et Exploit. des systèmes
  > Technology Services | Services technologiques
  > Informatics Branch | Direction de l'informatique
  > Maritimes Region, DFO | Région des Maritimes, MPO
  >
  > E-Mail: boivinp_at_mar.dfo-mpo.gc.ca
  > <mailto:boivinp_at_mar.dfo-mpo.gc.ca>
  >
  > --
  > Author: Boivin, Patrice J
  > INET: BoivinP_at_mar.dfo-mpo.gc.ca
  >
  > Fat City Network Services -- (858) 538-5051 FAX: (858)
  > 538-5051
  > San Diego, California -- Public Internet access /
  > Mailing Lists
  >
  

  ---

  
  > 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).

  ---

Gaja Krishna Vaidyanatha
Director, I-O Management Products
Quest Software Inc.
(972)-304-1170
gajav_at_yahoo.com

"Opinions and views expressed are my own and not of Quest"

---

Received on Wed Jul 12 2000 - 12:46:01 CDT