Solaris on UFS file systems may cause aync IO to be slow. Even though you have set it to "forcedirectio"
In that case it is not enough just to turn off disk_async_io. You might want to try "_lgwr_async_io"=FALSE
On 1/31/06, Jaco Polet <jaco.polet@(protected)> wrote: > > Anand > > Thanks for your feedback. > My answers are inline > I will test with a 2Mb log buffer > > Jaco > > > 2006/1/31, Anand Rao <panandrao@(protected)>: > > > > *Hi, > > > > what was the size of your log_buffer and what is the current size? you > > are only inserting about 32000 rows, so it shouldn't be a big deal. how > > frequent is this data load job? * > > > We doubled the log buffer from 0.5 Mb to 1Mb. No effect. > We only copied the data from one table to another. As the log shows it's > only 32000 rows stored in +/- 10 Mb. > > *'log buffer space' waits are generally caused due to a small log buffer > relative to the amount of redo you are generating. > > there is no direct reason why db_writer_processes has to be > increased.....for 'log buffer space' waits. If you have configured AsyncIO on the OS side, then you should use disk_ > async_io = TRUE. Are you using RAW volumes for your redologs and > datafiles? else, it doesn't make sense to turn on async IO. > * > ** > We changed disk_async_io to FALSE, set db_writer_processes to 2. We are > not using > > *I can't remember too many problems with Solaris 8 Async driver, but you > are better of checking it with your sysadmin as well searching Metalink for > any async IO related patches. > > If you are using UFS or VxFS for redo and datafiles, then what you need to > do is reduce your Filesystem Buffer Cache. Are you using Direct IO or some > new version of this 'near RAW performance Filesystem booster package' ? you > need to consider all these filesytem config options before deciding to use > Async IO. > * > We use Direct IO > * * > *When changed **you need to see if your Disk IO thoroughput is able to > handle the workload generated. you may have very few and slow disks causing > LGWR to be slow writing data from the log buffer to the redolog files. ** > > use a 1MB log buffer and see what you get. it all depends on your redo > rate. > > regards > anand * > > > > > > > On 31/01/06, Jaco Polet < jaco.polet@(protected) > wrote: > > > > > > Hello, > > > > > > I have a general issue with loading data into a Oracle 9.2.0.4database on solaris 8. I reduced it down to a simple statement which takes 1 > > > second on my PC (oracle 9.2.0.6) but takes 20 seconds on the 9.2.0 .4database on solaris. > > > I traced it and the trace showed a 19 second wait on log buffer space. > > > (see trace output) > > > I increased the value of log_buffer and set disk_async_io=FALSE > > > (increasing the db_writers) but this didn't make a difference > > > On metalink I found note 263652.1 saying that this was a general issue > > > on all platforms which could be fixed by switching on write cache. (It > > > doesn't state that is is fixed in a next version) > > > It sounds strange to me that Oracle recommends to write the redo log > > > to cache. Is this correct and if so how do I switch this on (the redologs > > > are on local disks) ? I am also surprised that when searching the internet I > > > don't get any hits on this although it looks like a general issue. What am I > > > missing? > > > > > > Hope someone can help me with this... Jaco Polet > > > > > > The formatted trace: > > > > > > insert into t_jpo select * from test_tabel > > > > > > call count cpu elapsed disk query > > > current rows > > > -- ---- -- --- -- ----- -- ---- -- -- ---- -- -- ---- -- -- ---- -- > > > -- ---- -- > > > Parse 1 0.01 0.00 0 0 > > > 0 0 > > > Execute 1 0.90 20.08 0 2920 > > > 7664 32114 > > > Fetch 0 0.00 0.00 0 0 > > > 0 0 > > > -- ---- -- --- -- ----- -- ---- -- -- ---- -- -- ---- -- -- ---- -- > > > -- ---- -- > > > total 2 0.91 20.09 0 2920 > > > 7664 32114 > > > > > > Misses in library cache during parse: 1 > > > Optimizer goal: CHOOSE > > > Parsing user id: 1340 (PERF) > > > > > > Rows Row Source Operation > > > -- ---- -- ---- ---- ---- ---- ---- ---- ---- ---- ---- --- > > > 32114 TABLE ACCESS FULL TEST_TABEL (cr=1182 r=0 w=0 time=96619 us) > > > > > > error during execute of EXPLAIN PLAN statement > > > ORA-00942 (See ORA-00942.ora-code.com): table or view does not exist > > > > > > parse error offset: 109 > > > > > > Elapsed times include waiting on following events: > > > Event waited on Times Max. Wait Total > > > Waited > > > -- ---- ---- ---- ---- ---- ---- ---- -- Waited -- ---- -- > > > -- ---- ---- > > > log buffer space 43 1.00 > > > 19.18 > > > SQL*Net message to client 1 0.00 > > > 0.00 > > > SQL*Net message from client 1 2.49 > > > 2.49 > > > ************************************************************************* ******* > > > > > > > > > > >
Solaris on <span id="__firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color: black; display: inline; font-size: inherit;" >UFS</span> file systems may cause aync IO to be slow. Even though you have set it to "forcedirectio" <br><br>In that case it is not enough just to turn off disk_async_io. You might want to try "_lgwr_<span id="__firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color: black; display: inline; font-size: inherit;"> async</span>_io"=FALSE<br><br><br><div><span class="gmail_quote">On 1/31 /06, <b class="gmail_sendername">Jaco Polet</b> <<a href="mailto:jaco.polet @(protected)">jaco.polet@(protected)</a>> wrote:</span><blockquote class="gmail _quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0 .8ex; padding-left: 1ex;"> <div>Anand</div> <div> </div> <div>Thanks for your feedback.</div> <div>My answers are inline</div> <div>I will test with a 2Mb log buffer</div> <div> </div> <div>Jaco</div> <div><br> </div> <div><span class="gmail_quote">2006/1/31, Anand Rao <<a href="mailto :panandrao@(protected)" target="_blank" onclick="return top.js.OpenExtLink(window ,event,this)">panandrao@(protected)</a>>:</span> <span class="q"><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0px 0px 0px 0.8ex; padding-left: 1ex;"><em>Hi,<br> <br>what was the size of your log_buffer and what is the current size? you are only inserting about 32000 rows, so it shouldn't be a big deal. how frequent is this data load job? </em></blockquote> <div> </div></span> <div>We doubled the log buffer from 0.5 Mb to 1Mb. No effect. </div> <div>We only copied the data from one table to another. As the log shows it's only 32000 rows stored in +/- 10 Mb.</div><span class="q"> <div> </div> <div><em>'log buffer space' waits are generally caused due to a small log buffer relative to the amount of redo you are generating. <br><br>there is no direct reason why db_writer_processes has to be increased.....for 'log buffer space' waits. If you have configured <span id="__firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color: black; display: inline; font-size: inherit;">Async</span> IO on the OS side, then you should use disk_<span id="__firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color: black; display: inline; font-size: inherit;"> async</span>_io = TRUE. Are you using RAW volumes for your redologs and datafiles? else, it doesn't make sense to turn on <span id="__firefox-findbar -search-id" style="padding: 0pt; background-color: yellow; color: black; display : inline; font-size: inherit;"> async</span> IO. <br></em></div><em></em></span></div> <div> <div>We changed disk_<span id="__firefox-findbar-search-id" style="padding: 0pt ; background-color: yellow; color: black; display: inline; font-size: inherit;" >async</span>_io to FALSE, set db_writer_processes to 2. We are not using </div><span class="q"> <div><br><em>I can't remember too many problems with Solaris 8 <span id="_ _firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color : black; display: inline; font-size: inherit;">Async</span> driver, but you are better of checking it with your sysadmin as well searching Metalink for any <span id="__firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color: black; display: inline; font-size: inherit;">async</span> IO related patches.<br><br>If you are using UFS or VxFS for redo and datafiles, then what you need to do is reduce your Filesystem Buffer Cache. Are you using Direct IO or some new version of this 'near RAW performance Filesystem booster package' ? you need to consider all these filesytem config options before deciding to use <span id="__firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color: black; display: inline; font-size: inherit;">Async</span> IO. <br></em></div></span> <div>We use Direct IO</div> <div><em> </em></div> <div><em>When changed </em><em>you need to see if your Disk IO thoroughput is able to handle the workload generated. you may have very few and slow disks causing LGWR to be slow writing data from the log buffer to the redolog files. </em><div><span class="e" id="q_10920b270b8ba587_7"><em><br><br>use a 1MB log buffer and see what you get. it all depends on your redo rate. <br><br>regards <br><span>anand</span> </em></span></div></div></div><div><span class="e" id="q _10920b270b8ba587_9">
<div> <blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204 ); margin: 0px 0px 0px 0.8ex; padding-left: 1ex;"> <div><span><br><br> <div><span class="gmail_quote">On 31/01/06, <b class="gmail_sendername">Jaco Polet</b> <<a href="mailto:jaco.polet@(protected)" target="_blank" onclick= "return top.js.OpenExtLink(window,event,this)"> jaco.polet@(protected) </a>> wrote:</span> <blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204 ); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"><font size="2"> </font> <div>Hello,</div> <div> </div> <div>I have a general issue with loading data into a Oracle <a href="http://9.2 .0.4/" target="_blank" onclick="return top.js.OpenExtLink(window,event,this)">9 .2.0.4</a> database on solaris 8. I reduced it down to a simple statement which takes 1 second on my PC (oracle <a href="http://9.2.0.6/" target="_blank" onclick="return top.js.OpenExtLink (window,event,this)">9.2.0.6</a>) but takes 20 seconds on the <a href="http://9 .2.0.4/" target="_blank" onclick="return top.js.OpenExtLink(window,event,this)">
9.2.0.4</a> database on solaris.</div> <div>I traced it and the trace showed a 19 second wait on log buffer space. (see trace output)</div> <div>I increased the value of log_buffer and set disk_<span id="_ _firefox-findbar-search-id" style="padding: 0pt; background-color: yellow; color : black; display: inline; font-size: inherit;">async</span>_io=FALSE (increasing the db_writers) but this didn't make a difference </div> <div>On metalink I found note 263652.1 saying that this was a general issue on all platforms which could be fixed by switching on write cache. (It doesn't state that is is fixed in a next version)</div> <div>It sounds strange to me that Oracle recommends to write the redo log to cache. Is this correct and if so how do I switch this on (the redologs are on local disks) ? I am also surprised that when searching the internet I don't get any hits on this although it looks like a general issue. What am I missing? </div> <div> </div> <div>Hope someone can help me with this... Jaco Polet</div> <div> </div> <div>The formatted trace:</div> <div> </div> <div>insert into t_jpo select * from test_tabel</div> <p>call count cpu elapsed disk query current rows<br>-- ---- -- --- -- ----- -- ---- -- -- ---- -- -- ---- --- -- ---- -- -- ---- --<br>Parse 1 0.01 0.00 0 0 0 0 <br>Execute 1 0.90 20.08 0 2920 7664 32114<br>Fetch 0 0.00 0.00 0 0 0 0<br>-- ---- -- --- -- ----- -- ---- -- -- ------ - -- ---- -- -- ---- -- -- ---- -- <span></span><br>total 2 0.91 20.09 0 2920 7664 32114</p> <p>Misses in library cache during parse: 1<br>Optimizer goal: CHOOSE<br>Parsing user id: 1340 (PERF)</p> <p>Rows Row Source Operation<br>-- ---- -- ------ -- ---- ---- ---- ---- ---- ---- ---- ----<br> 32114 TABLE ACCESS FULL TEST_TABEL (cr=1182 r=0 w=0 time=96619 us)</p> <p>error during execute of EXPLAIN PLAN statement<br>ORA-00942 (See ORA-00942.ora-code.com): table or view does not exist</p> <p>parse error offset: 109</p> <p>Elapsed times include waiting on following events:<br> Event waited on Times Max. Wait Total Waited<br> -- ---- -- -- ---- ---- ---- ---- ---- -- Waited -- ---- -- -- ---- -- --<br> log buffer space 43   ; 1.00 19.18<br> SQL*Net message to client 1 0.00 0.00<br> SQL*Net message from client 1 2.49 2.49<br>*********************** ********************************************************* <br></p></blockquote></div><br></span></div></blockquote></div><br>
