<div>Joel,</div> <div> </div> <div> It is not using o_direct only if the coreutils package was not installed on the RH3.0 machine. (<A href="http://oss.oracle.com/projects/coreutils/dist/files/EL3/i386/coreutils-4.5.3-41.i386.rpm">coreutils-4.5.3-41.i386.rpm</A> ).</div> <div> </div> <div><A href="http://oss.oracle.com/projects/coreutils/files/">http://oss.oracle.com/projects/coreutils/files/</A></div> <div> </div> <div> If it is installed, then both tests are using O_DIRECT, and can be compared.</div> <div> </div> <div> I do not have both a OCFS and a OCFS2 environment to compare here, but I am perceiving too a very slow performance with copy operations on the OCFS2 volume, compared to what I was used to in OCFS.</div> <div><BR>Regards,</div> <div>Luis</div> <div><BR><B><I>Joel Becker <Joel.Becker@oracle.com></I></B> wrote:</div> <BLOCKQUOTE class=replbq style="PADDING-LEFT: 5px; MARGIN-LEFT:
5px; BORDER-LEFT: #1010ff 2px solid">On Tue, Jan 16, 2007 at 01:28:41AM -0800, GOKHAN wrote:<BR>> Hi everbody this is my first post,<BR>> I have two test server .(Both of them is idle)<BR>> db1 : RHEL4 OCFS2<BR>> db2 : RHEL3 OCFS<BR>> <BR>> I test the IO both of them<BR>> The result is below.<BR>> <BR>> db1(Time Spend)db2(Time Spend)OS Test Command<BR>> dd (1GB) (Yazma)0m0.796s0m18.420stime dd if=/dev/zero of=./sill.t bs=1M count=1000<BR>> dd (1GB) (Okuma)0m0.241s8m16.406stime dd of=/dev/zero if=./sill.t bs=1M count=1000<BR>> cp (1GB)0m0.986s7m32.452stime cp sill.t sill2.t<BR><BR>You are using dd(1), which does not use O_DIRECT. The original<BR>ocfs (on 2.4 kernels) does not really support buffered I/O well. What<BR>you are seeing is ocfs2 taking much better care of your buffered I/Os.<BR>They will be consistent across the cluster. In the ocfs case, you are<BR>caching a lot more because these safety precautions aren't taken.<BR>HOWEVER,
the most important factor is that you are not using<BR>O_DIRECT. When you actually run the database, you _will_ be using<BR>O_DIRECT (make sure to mount ocfs2 with '-o datavolume'). Without the<BR>OS caching in the way, both filesystems should run at the same speed.<BR>The upshot is that buffered I/O operations (such as plain dd(1))<BR>are often not good indicators of database speed.<BR><BR>Joel<BR><BR>-- <BR><BR>"To announce that there must be no criticism of them president, or<BR>that we are to stand by the president, right or wrong, is not only<BR>unpatriotic and servile, but is morally treasonable to the American<BR>public."<BR>- Theodore Roosevelt<BR><BR>Joel Becker<BR>Principal Software Developer<BR>Oracle<BR>E-mail: joel.becker@oracle.com<BR>Phone: (650) 506-8127<BR><BR>_______________________________________________<BR>Ocfs2-users mailing
list<BR>Ocfs2-users@oss.oracle.com<BR>http://oss.oracle.com/mailman/listinfo/ocfs2-users<BR></BLOCKQUOTE><BR><p> __________________________________________________<br>Do You Yahoo!?<br>Tired of spam? Yahoo! Mail has the best spam protection around <br>http://mail.yahoo.com